AVR ATmega keeps resetting while using printf before main loop

Question

I'm developing a C application using avr-libc on an AVR ATmega328P microcontroller. Since I don't have an ICE debugger for it, I followed these instructions and this tutorial for making the stdio.h functions such as printf able to use the hardware UART as stdout.

That works, and I can see the output on a PC terminal connected to my target board, but the strange thing is: When I have only one printf on main, but before the main loop something is causing the processor to reset, while if I have a printf only inside the main loop or before the main loop AND inside the loop it works fine. Something like this:

#include <stdio.h>

/* stream definitions for UART input/output */
FILE uart_output = FDEV_SETUP_STREAM(uart_drv_send_byte, NULL, _FDEV_SETUP_WRITE);
FILE uart_input = FDEV_SETUP_STREAM(NULL, uart_drv_read_byte, _FDEV_SETUP_READ);

int main() {
    /* Definition of stdout and stdin */
    stdout = &uart_output;
    stdin = &uart_input;

    /* Configures Timer1 for generating a compare interrupt each 1ms (1kHz) */
    timer_init()

    /* UART initialization */
    uart_drv_start(UBRRH_VALUE, UBRRL_VALUE, USE_2X, &PORTB, 2);

    /* Sets the sleep mode to idle */
    set_sleep_mode(SLEEP_MODE_IDLE);

    printf("START ");

    /* main loop */
    while(1) {
        printf("LOOP ");

        /* Sleeps so the main loop iterates only on interrupts (avoids busy loop) */
        sleep_mode();
    }
}

The code above produces the following output:

START LOOP LOOP LOOP LOOP LOOP LOOP ... LOOP

which is expected. If we comment the printf("START ") line it produces this:

LOOP LOOP LOOP LOOP LOOP LOOP LOOP ... LOOP

which is also fine. The problem is, if I don't have any printf inside the while loop, it goes like this:

START START START START START START ... START

That clearly shows the processor is being restarted, since the expected output would be just one START and nothing else while the infinite loop goes on being awaken only on the 1 kHz timer interrupts. Why is this happening? I should stress there's no watchdog timer configured (if there was, the cases where only LOOP is printed would be interrupted by a new START also).

Monitoring execution using GPIO pins

To try to get some insight into the situation, I turned GPIO pins ON and OFF around the problematic print("START ") and sleep_mode in the main loop:

int main() {

    /* Irrelevant parts suppressed... */

    GPIO1_ON;
    printf("START ");
    GPIO1_OFF;

    /* Main loop */
    while(1) {

        /* Sleeps so the main loop iterates only on interrupts (avoids busy loop) */
        GPIO2_ON;
        sleep_mode();
        GPIO2_OFF;
    }
}

It turned out that GPIO1 stays ON for 132 µs (printf("START ") call time) and then OFF for 6.6 ms - roughly the time to transmit the six characters at 9600 bit/s - and GPIO2 toggles 12 times (six times two interrupts: the UART-ready-to-transmit interrupt and the UART-empty-data-register interrupt), showing sleep active for another 1.4 ms before GPIO1 goes ON again indicating a new printf("START ") - hence after reset. I'll probably have to check out the UART code, but I'm pretty sure the non-interrupt UART version also shows the same problem, and that doesn't explain either why having a printf inside the main loop works OK, without a reset happening (I would expect the reset would happen in any case should the UART code be faulty).

(SOLVED!): For completeness, The UART init and TX code is below**

This was my first attempt in writing an interrupt driven UART driver for the AVR, but one that could be used either on a RS-232 or a RS-485, which requires activating a TX_ENABLE pin while transmitting data. It turned out that, since I had to make the code useable either on ATmega328P or ATmega644, the interrupt vectors have different names, so I used a #define TX_VECTOR to assume the right name according to the processor used. In the process of making and testing the driver the choosing of "TX_VECTOR" for the UDRE data empty interrupt ended up masking the fact I hadn't defined the USART0_TX_vect yet (this was work in progress, I might not even need both anyway...)

Right now I just defined an empty interrupt service routine (ISR) for USART0_TX_vect and the thing doesn't reset anymore, showing @PeterGibson nailed it right on. Thanks a lot!

// Interrupt vectors for Atmega328P
#if defined(__AVR_ATmega328P__)
    #define RX_VECTOR USART_RX_vect
    #define TX_VECTOR USART_UDRE_vect
    // Interrupt vectors for Atmega644
#elif defined(__AVR_ATmega644P__)
    #define RX_VECTOR USART0_RX_vect
    #define TX_VECTOR USART0_UDRE_vect
#endif

ISR(TX_VECTOR)
{
    uint8_t byte;

    if (!ringbuffer_read_byte(&txrb, &byte)) {

        /* If RS-485 is enabled, sets TX_ENABLE high */
        if (TX_ENABLE_PORT)
            *TX_ENABLE_PORT |= _BV(TX_ENABLE_PIN);
        UDR0 = byte;
    }
    else {
        /* No more chars to be read from ringbuffer, disables empty
         * data register interrupt */
        UCSR0B &= ~_BV(UDRIE0);
    }

    /* If RS-485 mode is on and the interrupt was called with TXC0 set it
     * means transmission is over. TX_ENABLED should be cleared. */
    if ((TX_ENABLE_PORT) && (UCSR0A & _BV(TXC0) & _BV(UDR0))) {
        *TX_ENABLE_PORT &= ~_BV(TX_ENABLE_PIN);
        UCSR0B &= ~_BV(UDRIE0);
    }
}

void uart_drv_start(uint8_t ubrrh, uint8_t ubrrl, uint8_t use2x,
                    volatile uint8_t* rs485_tx_enable_io_port,
                    uint8_t rs485_tx_enable_io_pin)
{
    /* Initializes TX and RX ring buffers */
    ringbuffer_init(&txrb, &tx_buffer[0], UART_TX_BUFSIZE);
    ringbuffer_init(&rxrb, &rx_buffer[0], UART_RX_BUFSIZE);

    /* Disables UART */
    UCSR0B = 0x00;

    /* Initializes baud rate */
    UBRR0H = ubrrh;
    UBRR0L = ubrrl;
    if (use2x)
        UCSR0A |= _BV(U2X0);
    else
        UCSR0A &= ~_BV(U2X0);

    /* Configures async 8N1 operation */
    UCSR0C = _BV(UCSZ00) | _BV(UCSZ01);

    /* If a port was specified for a pin to be used as a RS-485 driver TX_ENABLE,
     * configures the pin as output and enables the TX data register empty
     * interrupt so it gets disabled in the end of transmission */
    if (rs485_tx_enable_io_port) {
        TX_ENABLE_PORT = rs485_tx_enable_io_port;
        TX_ENABLE_PIN = rs485_tx_enable_io_pin;
        /* Configures the RS-485 driver as an output (on the datasheet the data
         * direction register is always on the byte preceding the I/O port addr) */
        *(TX_ENABLE_PORT-1) |= _BV(TX_ENABLE_PIN);
        /* Clears TX_ENABLE pin (active high) */
        *TX_ENABLE_PORT &= ~_BV(TX_ENABLE_PIN);
        /* Enables end of transmission interrupt */
        UCSR0B = _BV(TXCIE0);
    }
    /* Enables receptor, transmitter and RX complete interrupts */
    UCSR0B |= _BV(RXEN0) | _BV(TXEN0) | _BV(RXCIE0);
}

FIXED UART CODE (NOW WORKING 100%!)

In order to help anyone interested or developing a similar interrupt driven UART driver for the AVR ATmega, here it goes the code with the problems above fixed and tested. Thanks to everyone who helped me spot the problem with the missing ISR!

// Interrupt vectors for Atmega328P
#if defined(__AVR_ATmega328P__)
    #define RX_BYTE_AVAILABLE USART_RX_vect
    #define TX_FRAME_ENDED USART_TX_vect
    #define TX_DATA_REGISTER_EMPTY USART_UDRE_vect
    // Interrupt vectors for Atmega644
#elif defined(__AVR_ATmega644P__)
    #define RX_BYTE_AVAILABLE USART0_RX_vect
    #define TX_FRAME_ENDED USART0_TX_vect
    #define TX_DATA_REGISTER_EMPTY USART0_UDRE_vect
#endif

/* I/O port containing the pin to be used as TX_ENABLE for the RS-485 driver */
static volatile uint8_t* TX_ENABLE_PORT = NULL;

/** Pin from the I/O port to be used as TX_ENABLE for the RS-485 driver */
static volatile uint8_t TX_ENABLE_PIN = 0;

ISR(RX_BYTE_AVAILABLE)
{
    // Read the status and RX registers.
    uint8_t status = UCSR0A;
    // Framing error - treat as EOF.
    if (status & _BV(FE0)) {
        /* TODO: increment statistics */
    }
    // Overrun or parity error.
    if (status & (_BV(DOR0) | _BV(UPE0))) {
        /* TODO: increment statistics */
    }
    ringbuffer_write_byte(&rxrb, UDR0);
}

ISR(TX_FRAME_ENDED)
{
    /* The end of frame interrupt will be enabled only when in RS-485 mode, so
     * there is no need to test, just turn off the TX_ENABLE pin */
    *TX_ENABLE_PORT &= ~_BV(TX_ENABLE_PIN);
}

ISR(TX_DATA_REGISTER_EMPTY)
{
    uint8_t byte;

    if (!ringbuffer_read_byte(&txrb, &byte)) {
        /* If RS-485 is enabled, sets TX_ENABLE high */
        if (TX_ENABLE_PORT)
            *TX_ENABLE_PORT |= _BV(TX_ENABLE_PIN);
        UDR0 = byte;
    }
    else {
        /* No more chars to be read from ringbuffer, disables empty
         * data register interrupt */
        UCSR0B &= ~_BV(UDRIE0);
    }
}

void uart_drv_start(uint8_t ubrrh, uint8_t ubrrl, uint8_t use2x,
                    volatile uint8_t* rs485_tx_enable_io_port,
                    uint8_t rs485_tx_enable_io_pin)
{
    /* Initializes TX and RX ring buffers */
    ringbuffer_init(&txrb, &tx_buffer[0], UART_TX_BUFSIZE);
    ringbuffer_init(&rxrb, &rx_buffer[0], UART_RX_BUFSIZE);

    cli();

    /* Disables UART */
    UCSR0B = 0x00;

    /* Initializes baud rate */
    UBRR0H = ubrrh;
    UBRR0L = ubrrl;
    if (use2x)
        UCSR0A |= _BV(U2X0);
    else
        UCSR0A &= ~_BV(U2X0);

    /* Configures async 8N1 operation */
    UCSR0C = _BV(UCSZ00) | _BV(UCSZ01);

    /* If a port was specified for a pin to be used as a RS-485 driver TX_ENABLE,
     * configures the pin as output and enables the TX data register empty
     * interrupt so it gets disabled in the end of transmission */
    if (rs485_tx_enable_io_port) {
        TX_ENABLE_PORT = rs485_tx_enable_io_port;
        TX_ENABLE_PIN = rs485_tx_enable_io_pin;

        /* Configures the RS-485 driver as an output (on the datasheet the data
         * direction register is always on the byte preceding the I/O port addr) */
         *(TX_ENABLE_PORT-1) |= _BV(TX_ENABLE_PIN);

        /* Clears TX_ENABLE pin (active high) */
        *TX_ENABLE_PORT &= ~_BV(TX_ENABLE_PIN);

        /* Enables end of transmission interrupt */
        UCSR0B = _BV(TXCIE0);
    }
    /*  Enables receptor, transmitter and RX complete interrupts */
    UCSR0B |= _BV(RXEN0) | _BV(TXEN0) | _BV(RXCIE0);

    sei();
}

void uart_drv_send_byte(uint8_t byte, FILE *stream)
{
    if (byte == '\n') {
        uart_drv_send_byte('\r', stream);
    }
    uint8_t sreg = SREG;
    cli();

    /* Write byte to the ring buffer, blocking while it is full */
    while(ringbuffer_write_byte(&txrb, byte)) {
        /* Enable interrupts to allow emptying a full buffer */
        SREG = sreg;
        _NOP();
        sreg = SREG;
        cli();
    }

    /* Enables empty data register interrupt */
    UCSR0B |= _BV(UDRIE0);
    SREG = sreg;
}

uint8_t uart_drv_read_byte(FILE *stream)
{
    uint8_t byte;
    uint8_t sreg = SREG;
    cli();
    ringbuffer_read_byte(&rxrb, &byte);
    SREG = sreg;
    return byte;
}

Without saying the problem is a watchdog, I'm not sure I agree with your logic on why that could not be the problem. Where's the implementation of sleep_mode()? Although you don't have a debugger, perhaps you should objdump the output and see what the actual instructions look like in each case. Also, you could try something like toggling a GPIO in the loop as an alternative to the printf() and thus see if it has anything to do with the serial port, or if it's about having an empty loop, or? — Chris Stratton
@Chris, the implementation of sleep_mode is the one from avr-libc, so I really didn't look much into it, just used it. I did have something else on the loop besides it, I just commented everything out so it would be easier to pinpoint the problem, but it also happens when something else is done before sleep. I'm just about to try setting a GPIO before and after sleep so we can have some more insight into it. — Claudio
What happens when there is no more data to send? The error could be related to an empty output buffer. — D Krueger
Where is uart_drv_start defined? It doesn't seem to be in the std libs — Peter Gibson
@PeterGibson, that function is not in the libs indeed. I initially didn't suspect about it because the non-interrupt version also showed the same problem, but now I see the problem might be just there anyway. — Claudio

Peter Gibson Peter Gibson · Accepted Answer · 2014-02-05T02:12:50

You've possibly enabled the UDRE (Uart Data Register Empty) interrupt and not set a vector for it, so when the interrupt triggers the processor resets (according to the defaults). When printf is called continuously in the main loop, this interrupt is never triggered.

From the docs

Catch-all interrupt vector

If an unexpected interrupt occurs (interrupt is enabled and no handler is installed, which usually indicates a bug), then the default action is to reset the device by jumping to the reset vector. You can override this by supplying a function named BADISR_vect which should be defined with ISR() as such. (The name BADISR_vect is actually an alias for __vector_default. The latter must be used inside assembly code in case is not included.)

AVR ATmega keeps resetting while using printf before main loop

2 Answers