GPIO Interrupt latency

[almog1]

Hey,
I am trying to use GPIO as input trigger to my system.
I am using shell command before I am running my code:

echo in          >/sys/class/gpio/gpio443/direction
echo falling  >/sys/class/gpio/gpio443/edge

and also a C code:

void gpio_test()
{
    if ((fd = open("/sys/class/gpio/gpio443/value", O_RDWR)) < 0)
    {
        fprintf(stderr, "Failed, gpio %d not exported.\n", gpioPin);
        return -1;
    }

    pfd.fd = fd;

    pfd.events = POLLPRI;

    lseek(fd, 0, SEEK_SET); /* consume any prior interrupt */
    read(fd, buf, sizeof(buf));
    rc = poll(&pfd, 1, timeoutMS); /* wait for interrupt */


    if (rc < 0)
        strerror(errno);

    lseek(fd, 0, SEEK_SET); /* consume interrupt */
    read(fd, buf, sizeof(buf));
}

I see that the interrupt arrived with ~15 ms latency after the GPIO fall down.
The program runs in user space.

There is any way to decrease the latency?

[JerryChang]

hello almog1,

could you please check Topic 192147 for the steps to improve GPIO interrupt performance.
thanks

[almog1]

Hey Jerry,
I have tried to use attached commands:

sudo su
cat /sys/kernel/debug/bpmp/debug/clk/axi_cbb/max_rate # 408000000

cat /sys/kernel/debug/bpmp/debug/clk/axi_cbb/max_rate > /sys/kernel/debug/bpmp/debug/clk/axi_cbb/rate
echo 1 > /sys/kernel/debug/bpmp/debug/clk/axi_cbb/mrq_rate_locked 

cat /sys/kernel/debug/bpmp/debug/clk/axi_cbb/rate # 408000000
cat /sys/kernel/debug/bpmp/debug/clk/axi_cbb/max_rate # 408000000
cat /sys/kernel/debug/bpmp/debug/clk/axi_cbb/mrq_rate_locked # 1

I am still get the interrupt with ~15ms latency.

root@nvidia:/home/nvidia# cat /proc/interrupts | grep gpiolib
 353:       8767          0          0          0          0          0          0          0  tegra-gpio  155 Edge      gpiolib

[JerryChang]

@almog1

I am still get the interrupt with ~15ms latency.

may I know how you evaluate 15ms latency,
are you probing the signal or adding kernel logs to check the timestamp?

[almog1]

Hey Jerry,
You right, Actually I have checked it via log file.
In case of checking it with oscilloscope it actually works fine.
Thank you

[JerryChang]

There is no update from you for a period, assuming this is not an issue any more.
Hence we are closing this topic. If need further support, please open a new one.
Thanks

hello almog1,

would you please use $ nice to set higher process priority to confirm the behavior?
thanks