Firmware hangup in `i2c_stop()` and other I2C functions

[maxgerhardt]

I observed it multiple times in a hardware configuration with a SSD1306 I2C OLED display that reflashing the firmware at specific points would cause the new firmware to hangup in the i2c_stop() function (as I observed in the debugger).

https://github.com/CommunityGD32Cores/GD32Core-New/blob/88d1cd79e196e13deaa67faeb7aea008f10157fd/libraries/Wire/src/utility/twi.c#L144-L155

This code does not have timeouts (as opposed to other I2C functions) and in case the I2C peripherhal wants to assert a STOP condition but the bus just doesn't let it be driven that way (some other device is interfering?) this hangs up forever.

This should be investigated and fixed.

[maxgerhardt]

A timeout has been added in a72b50c and 51a22bf, further improvements in PR #45 #46

a backtrace of a lockup is also e.g.

Program received signal SIGINT, Interrupt.
0x08001948 in i2c_byte_write (obj=0x20000104 <Wire+20>, data=<optimized out>)
    at C:\Users\Max\.platformio\packages\framework-arduinogd32\libraries\Wire\src\utility\twi.c:157
157         while(((i2c_flag_get(obj_s->i2c, I2C_FLAG_TBE)) == RESET) &&
(gdb) where
#0  0x08001948 in i2c_byte_write (obj=0x20000104 <Wire+20>,
    data=<optimized out>)
    at C:\Users\Max\.platformio\packages\framework-arduinogd32\libraries\Wire\src\utility\twi.c:157
#1  i2c_master_transmit (obj=obj@entry=0x20000104 <Wire+20>,
    address=<optimized out>, data=0x2000014c <TwoWire::_tx_buffer> "",
    length=<optimized out>, stop=1 '<error reading variable>)
    at C:\Users\Max\.platformio\packages\framework-arduinogd32\libraries\Wire\src\utility\twi.c:241
#2  0x08001648 in TwoWire::endTransmission (this=0x200000f0 <Wire>,
    sendStop=sendStop@entry=1 '<error reading variable>)
    at C:\Users\Max\.platformio\packages\framework-arduinogd32\libraries\Wire\src\Wire.cpp:191
#3  0x08001666 in TwoWire::endTransmission (this=<optimized out>)
    at C:\Users\Max\.platformio\packages\framework-arduinogd32\libraries\Wire\src\Wire.cpp:207
#4  0x08002dfe in Adafruit_SSD1306::ssd1306_commandList (
    this=this@entry=0x20000084 <display>, c=<optimized out>,
    c@entry=0x8006f3a <Adafruit_SSD1306::begin(unsigned char, unsigned char, bool, bool)::init1> <error reading variable>, n=<optimized out>,
    n@entry=4 '<error reading variable>)
    at .pio\libdeps\genericGD32F303CC\Adafruit SSD1306\Adafruit_SSD1306.cpp:402
#5  0x08002f98 in Adafruit_SSD1306::begin (
    this=this@entry=0x20000084 <display>,
    vcs=vcs@entry=2 '<error reading variable>,
    addr=addr@entry=60 '<error reading variable>, reset=reset@entry=true,
    periphBegin=periphBegin@entry=true)
    at .pio\libdeps\genericGD32F303CC\Adafruit SSD1306\Adafruit_SSD1306.cpp:537
#6  0x08001402 in setup () at src\main.cpp:82
#7  0x08004cd6 in main ()
    at C:\Users\Max\.platformio\packages\framework-arduinogd32\cores\arduino\main.cpp:51

So really the lockup can happen at any time.

• the timeout value is extremely large
• there is no error handling if timeouts fail -- other libraries try to do I2C bus resets then
  • see https://www.i2c-bus.org/i2c-primer/analysing-obscure-problems/blocked-bus/
  • Wire library lockups recoverable with timeout but without hardware reset arduino/ArduinoCore-avr#326

The Wire library needs an overhaul regarding error / timeout checking and I2C bus resetting.

I can reproduce most lock-ups very determinstically when I flash the I2C OLED benchmark demo and reflash it in the middle of when it's writing to the screen.