How to use ILA cross trigger for AXI?

Question

I need the Xilinx System ILA IP core to debug the AXI bus of a design. So I´ve created a more simplier design to playing around with the ILA core and to understand how it works.

With the following software

#include "xgpio.h"
#include "xparameters.h"

int main()
{
    XGpio Gpio;

    u32 Status = XGpio_Initialize(&Gpio, XPAR_LED_GPIO_DEVICE_ID);
    if(Status != XST_SUCCESS)
    {
        xil_printf("Gpio Initialization failed!\r\n");
        return XST_FAILURE;
    }

    while(1)
    {
        u32 Input = XGpio_DiscreteRead(&Gpio, 2);
        XGpio_DiscreteWrite(&Gpio, 1, Input);
        for(u32 Delay = 0; Delay < 0x10000; Delay++);
    }

    return 0;
}

I also enable the cross trigger interface from the processing system to the ILA in the debug configuration.

I set the my breakpoint to the line XGpio_DiscreteWrite(&Gpio, 1, Input); to capture the AXI communication from the processing system to the gpio core. Now I run the debug configuration and open the ILA with the following settings (default settings):

After continuing my application the ILA sample the communication with the gpio core.

But the core doesn´t sample the second transmission (the transmission for the XGpio_DiscreteWrite(&Gpio, 1, Input) line).

I´m quite sure that I´m using the ILA wrong. So how can I improve the usage of the ILA? What is the meaning of Number of windows in the ILA configuration?

Answer 1

The ILA grabs a single window of data for each trigger that it sees. So once the first one has happened, it will not collect any more data after the end of that window. If the second transaction does not come along within that window then you will not see it.

If you set the number of windows to '2' (or more), you will see both transactions as connected to separate triggers. However, now you have no idea how big the time difference between the two triggers was. And you get less data in each window :(

Another way to use the tool which is sometimes handy is to set the window length to a single sample and have (for example) 1024 windows of 1 sample each. Then set the trigger to be on any interesting signal change. You will then see all the changes in these signals. You lose all the timing information, but you have the ability to see lots of transactions, which is sometimes the more useful thing to see.