ESP32C3M1: Serial.print() calls when not connected to USB cause delays.

[mahboud]

Board

ESP32C3M1

Device Description

Custom board:

Hardware Configuration

GPIO 18 & 19 are connected to USB. USB is used to program the device and to monitor Serial.print output.

Version

v2.0.3

IDE Name

Arduino

Operating System

MacOS 10.15.7

Flash frequency

40Mhz

PSRAM enabled

yes

Upload speed

115200

Description

USB is used to program the device and to monitor Serial.print output. However, when the device is unplugged, and the Serial.print() lines are left in the code, they cause delays of a second or so. Plugging back in will make the delays go away. If I remove the Serial.print() lines, it makes the delays disappear too.

Sketch

#include <Wire.h>
#include <Adafruit_NeoPixel.h>

#define LED_PIN 6
#define NEOPIXEL_PIN LED_PIN
Adafruit_NeoPixel pixels(1, NEOPIXEL_PIN, NEO_GRB + NEO_KHZ800);

void setup() {
  // put your setup code here, to run once:
  Serial.begin(115200);

}

void loop() {
  // put your main code here, to run repeatedly:

  pixels.begin();  // INITIALIZE NeoPixel strip object (REQUIRED)
  
  Serial.print("Red... ");
  // drawMessage("Test Red LED");
  pixels.setPixelColor(0, pixels.Color(64, 0, 0));
  pixels.show();
  delay(500);

    Serial.print("Green... ");
// drawMessage("Test Green LED");
  pixels.setPixelColor(0, pixels.Color(0, 64, 0));
  pixels.show();
  delay(500);

    Serial.print("Blue... ");
// drawMessage("Test Blue LED");
  pixels.setPixelColor(0, pixels.Color(0, 0, 64));
  pixels.show();
  delay(500);
  
    Serial.print("White... ");
// drawMessage("Test Whitish LED");
  pixels.setPixelColor(0, pixels.Color(64, 64, 64));
  pixels.show();
  delay(500);

  Serial.print("Off... ");
  pixels.clear();
  pixels.show();
  delay(500);
  Serial.println("Done. Looping");
}```

Debug Message

There is no debug or error message.

Other Steps to Reproduce

Run the sketch. While the device is plugged into a computer with Arduino, the lights flash faster. When you unplug the USB cable the lights slow down.

With more or longer Serial.print output, the delay can be more than a second.

That same code above is running on the two boards in this video. When you plug the USB in, you can see that the flash rate increases on the one plugged in.

IMG_7923.mov

I have checked existing issues, online documentation and the Troubleshooting Guide

• I confirm I have checked existing issues, online documentation and Troubleshooting guide.

[SuGlider]

Issue confirmed.

I Can see the issue using a ESP32-C3 DevKitM-1 v1.0.
I connected both USB CDC and UART ports to the PC.
Arduino IDE iis configured with USB CDC On Boot: "Enabled"
It's possible to see the output of Serial0 in the UART.

Note:

I see that the issue occurs even with the USB plugged in, but the IDE Monitor is on the UART port instrad of on the CDC, as we can see in the UART output. When the USB CDC Port isn't open, the issue occurs, independently is the USB is plugged or not.
The time spent on a loop goes from 2.5 seconds to 3.9 seconds after a few interactions.

If the skecth is uploaded with USB CDC On Boot: "Disabled", the issue doesn't happen, and the loop takes always 2.5 seconds.

This makes it clear that USB CDC has some issue when its buffer gets full and it is still closed or unplugged.

This is the sketch used:

#include <Wire.h>
#include <Adafruit_NeoPixel.h>

#define LED_PIN 8   //6
#define NEOPIXEL_PIN LED_PIN
Adafruit_NeoPixel pixels(1, NEOPIXEL_PIN, NEO_GRB + NEO_KHZ800);

void setup() {
  // put your setup code here, to run once:
  Serial.begin(115200);
  Serial0.begin(115200);

}

void loop() {
  // put your main code here, to run repeatedly:

  pixels.begin();  // INITIALIZE NeoPixel strip object (REQUIRED)

  uint32_t now = millis();

  Serial.print("Red... ");
  // drawMessage("Test Red LED");
  pixels.setPixelColor(0, pixels.Color(64, 0, 0));
  pixels.show();
  delay(500);

  Serial.print("Green... ");
  // drawMessage("Test Green LED");
  pixels.setPixelColor(0, pixels.Color(0, 64, 0));
  pixels.show();
  delay(500);

  Serial.print("Blue... ");
  // drawMessage("Test Blue LED");
  pixels.setPixelColor(0, pixels.Color(0, 0, 64));
  pixels.show();
  delay(500);

  Serial.print("White... ");
  // drawMessage("Test Whitish LED");
  pixels.setPixelColor(0, pixels.Color(64, 64, 64));
  pixels.show();
  delay(500);

  Serial.print("Off... ");
  pixels.clear();
  pixels.show();
  delay(500);
  Serial.println("Done. Looping");

  Serial0.printf("\n========================\nLoop time = %d\n", millis()-now);
}

UART OUPUT:

ESP-ROM:esp32c3-api1-20210207
Build:Feb  7 2021
rst:0x1 (POWERON),boot:0xc (SPI_FAST_FLASH_BOOT)
SPIWP:0xee
mode:DOUT, clock div:1
load:0x3fcd6100,len:0x38c
load:0x403ce000,len:0x6ac
load:0x403d0000,len:0x24e4
SHA-256 comparison failed:
Calculated: 536dfe3b000041806b8028e5cb5125cd46736c37326abcaa489747c6c6fe4906
Expected: ffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff
Attempting to boot anyway...
entry 0x403ce000

========================
Loop time = 2508

========================
Loop time = 2501

========================
Loop time = 2501

========================
Loop time = 2501

========================
Loop time = 2901

========================
Loop time = 3901

========================
Loop time = 3901

========================
Loop time = 3901

========================
Loop time = 3901

[markxoe]

The delay comes from the timeout here (I think)

To (temporarily) solve the problem we could use a check whether the USB is connected or not before every write to the USB...

To fix it even more temporarily you could use Serial.setTxTimeoutMs(10); to lower the blocking time

[egnor]

What do we think a good solution would be?

In my own code I attempted using

if (Serial) Serial.printf("blah blah\n");

which I thought would prevent the timeout (with the possible exception of a print in progress during disconnect, and/or log statements I have no control over). Reading HWCDC.cpp, it does seem like operator bool() is supposed to return whether the port is connected-- it returns initial_empty, which is set to true when the ESP-to-host buffer is empty and writable, and set to false when there's a bus reset. However, this didn't work... maybe those conditions don't exactly align with USB connected/disconnected (and the CDC device being opened/closed by the host).

In general doing a good race-free job of quickly skipping writes when nothing is reading them (but blocking when something is) seems a bit tricky?