Running oneshot or periodic timers

Applications may likely need to run timed loops waiting for high-precision clock events, either once or according to a recurring period. The EVL core gives them timers for this purpose, which can be obtained on any existing EVL clock. EVL timers support synchronous delivery, there is no asynchronous notification mechanism of clock events. This means that a client thread must explicitly wait for the next expiry by either reading or polling for such event. Would the application choose to poll for clock events on a timer, it would still have to collect these events once they have occurred by reading the timer file descriptor.

Each EVL timer is referred to by a file descriptor. The EVL timer API mimics the timerfd API for the most part, which consists of:

  1. Creating a timer.
  2. Setting up a timeout date, either oneshot or recurring.
  3. Waiting for the (next) timeout to elapse by calling oob_read() for the timer file descriptor. The value collected by reading a timer is the number of elapsed ticks since the last readout. Therefore, any value greater than 1 would denote an overrun condition.

A simple periodic loop using an EVL timer

	#include <evl/thread.h>
	#include <evl/timer.h>
	#include <evl/clock.h>
	#include <evl/proxy.h>

	void timespec_add_ns(struct timespec *__restrict r,
	     		     const struct timespec *__restrict t,
			     long ns)
		long s, rem;

		s = ns / 1000000000;
		rem = ns - s * 1000000000;
		r->tv_sec = t->tv_sec + s;
		r->tv_nsec = t->tv_nsec + rem;
		if (r->tv_nsec >= 1000000000) {
		     r->tv_nsec -= 1000000000;

	int main(int argc, char *argv[])
		struct itimerspec value, ovalue;
		int tfd, tmfd, ret, n = 0;
		struct timespec now;
		__u64 ticks;

		/* Attach to the core. */
		tfd = evl_attach_self("periodic-timer:%d", getpid());

		/* Create a timer on the built-in monotonic clock. */
		tmfd = evl_new_timer(EVL_CLOCK_MONOTONIC);

		/* Set up a 1 Hz periodic timer. */
		ret = evl_read_clock(EVL_CLOCK_MONOTONIC, &now);
		/* EVL always uses absolute timeouts, add 1s to the current date */
		timespec_add_ns(&value.it_value, &now, 1000000000ULL);
		value.it_interval.tv_sec = 1;
		value.it_interval.tv_nsec = 0;
		ret = evl_set_timer(tmfd, &value, &ovalue);

		for (;;) {
		    /* Wait for the next tick to be notified. */
		    ret = oob_read(tmfd, &ticks, sizeof(ticks));
		    if (ticks > 1) {
		       	      fprintf(stder, "timer overrun! %lld ticks late\n",
			      	      ticks - 1);
		    evl_printf("TICKED, loops=%d\n", n++);

		/* Disable the timer (not required if closing). */
		value.it_interval.tv_sec = 0;
		value.it_interval.tv_nsec = 0;
		ret = evl_set_timer(tmfd, &value, NULL);

		return 0;

Timer services

int evl_new_timer(int clockfd)

This call creates an EVL timer on the clock referred to by clockfd, returning a file descriptor representing the new object upon success. There is no arbitrary limit on the number of timers an application can create, which is only limited to the available system resources.

  • clockfd

    The file descriptor of the reference clock the new timer should be synchronized on. clockfd refer to any valid clock element known from the EVL core, including one of the pre-defined clocks.

  • evl_new_timer() returns the file descriptor of the newly created timer on success. Otherwise, a negated error code is returned:

    • -EMFILE The per-process limit on the number of open file descriptors has been reached.

    • -ENFILE The system-wide limit on the total number of open files has been reached.

    • -ENOMEM No memory available.

    • -ENXIO The EVL library is not initialized for the current process. Such initialization happens implicitly when evl_attach_self() is called by any thread of your process, or by explicitly calling evl_init(). You have to bootstrap the library services in a way or another before creating an EVL timer.

    int evl_set_timer(int efd, struct itimerspec *value, struct itimerspec *ovalue)

    You should use this call to arm or disarm an EVL timer. It sets the next expiration date and reload value for the timer referred to by efd. If ovalue is not NULL, the current expiration date and reload value are stored at this location prior to updating them, as with evl_get_timer.

  • efd

    A file descriptor referring to the EVL timer to set up.

  • value

    A pointer to the interval timer specification structure which contains the new absolute expiration date and reload value. If both value->it_value.tv_nsec and value->it_value.tv_sec are zero, the timer is disarmed and value->it_interval is ignored. Otherwise, the timer is program to timeout at the specified date mentioned in value->it_value. If at least on field of value->it_interval is non-zero, the EVL will reprogram the timer after each timeout according to this period automatically.

  • ovalue

    If non-NULL, the previous expiration date and reload value of the timer are copied to this memory location.

  • Zero is returned on success. Otherwise, a negated error code is returned:

    • -EBADF efd is not a valid file descriptor referring to an EVL timer.

    • -EINVAL some field in *value is not valid, likely it_value.tv_nsec and/or it_interval.tv_nsec are out of the valid range, which is between 0 and 1e9 - 1.

    • -EFAULT either value or ovalue (if non-NULL) point to invalid memory.

    int evl_get_timer(int efd, struct itimerspec *value)

    Retrieve the current timeout and reload values of an EVL timer.

  • efd

    A file descriptor referring to the EVL timer to query.

  • value

    The current expiration date and reload value of the timer are copied to this memory location. The expiration date is always given in absolute form based on the reference clock for the timer.

  • Zero is returned on success. Otherwise, a negated error code is returned:

    • -EBADF efd is not a valid file descriptor referring to an EVL timer.

    • -EFAULT if value points to invalid memory.

    Events pollable from a timer file descriptor

    The evl_poll() interface can monitor the following events occurring on a timer file descriptor:

    • POLLIN and POLLRDNORM are set whenever a timeout event has fired for the timer, which is yet to be collected by the parent process. Timer events should be awaited for and collected by a call to oob_read().

    Last modified: Thu, 03 Aug 2023 11:30:05 +0200