[RP2040] Serial2.begin crashes when RX Data incomming

[savejeff]

I work on a project with the RP2040 (Pico) where both Serial Ports are needed. The second HardwareSerial _UART2_ does not work correctly.

This simple example:

UART Serial2(8,9);

void setup() {
	Serial.begin(115200);
	Serial2.begin(9600);

	delay(3000);
	Serial.println("Start");
}

void loop() {
	if (Serial2.available()) {
		Serial.write(Serial2.read());
	}
}

does not reach the Serial.println("Start"); line if there is Data incoming over the RX pin of the Serial2.
it initializes fine and the forwarding works if I disconnect the RX pin during the setup and then connect it after Start is printed. Then RX data works no problem.

to Test this I connected a second Pico to generate some Serial Data with this sketch:

void setup() {

	Serial1.begin(9600);
}

void loop() {
	Serial1.println("abcdefghijklmnopqrstuvwxyz01234567890");
	delay(250);
}

I was able to fix this bug by increasing the rx_buffer size in Serial.h. if I change it to RingBufferN<(1 << 12)> rx_buffer;
This points to a buffer overflow.

it seems to be a problem with the UART::on_rx() function. If the RX buffer is full no bytes are read from _serial->obj. this a guess causes the system to stall. I was able to fix this by just dropping serial data if the RX buffer is full. here is a working code:

void UART::on_rx() {
#if defined(SERIAL_CDC)
	if (is_usb) {
		return;
	}
#endif

	while(_serial->obj->readable()) {
		char c;
		core_util_critical_section_enter();
		_serial->obj->read(&c, 1);
		if(rx_buffer.availableForStore())
			rx_buffer.store_char(c);
		core_util_critical_section_exit();
	}
}

Receiving from 2 Serials at the same time still seems to not work but at least it does not crash anymore

[LeoDJ]

This also happens on Serial1. I stumbled over this issue too recently.
If you don't read from the Serial buffer, the Serial1 buffer runs full after exactly 256 bytes and the RP2040 will stall.

I tried to debug into the issue, it definitely hangs at _serial->obj->readable() in Serial.cpp:135 but couldn't find the implementation for UnbufferedSerial::readable() and thus was stumped why it would freeze.

But yes, simply dropping the incoming data if the buffer is full is the correct way imo and also how other implementations do it.

[facchinm]

Hi @LeoDJ, thanks for spotting this and for the workaround!
Would you mind submitting a PR so we can give the proper attribution to the fix?

[LeoDJ]

The workaround implementation was provided by @savejeff, not by me, so he should be the one to submit the PR ^^'

[savejeff]

Thx @LeoDJ, but I'm not that interested in the attribution.

I rather clearly documented the copy/paste fix. But as I stated, it's just a dirty fix, as receiving Data on both Serial1 and Serial2 simultaneously does not work even after the fix. (At the time of testing at least)

I'm not familiar with the mbed implementation but I feel it's something about there being only a single rx buffer for both serial or something like that.

[facchinm]

Merged as b14a86a .
I couldn't however reproduce the other issue (receiving data on both Seria1 and Serial2); the queues / buffer are totally independent so I don't see how it could interfere...
It would be nice if you could provide a minimal PoC that shows the issue on a single board