man timer_create

TIMER_CREATE(2)            Linux Programmer's Manual           TIMER_CREATE(2)
NAME
       timer_create - create a POSIX per-process timer
SYNOPSIS
       #include <signal.h>
       #include <time.h>
       int timer_create(clockid_t clockid, struct sigevent *sevp,
                        timer_t *timerid);
 
       Link with -lrt.
 
   Feature Test Macro Requirements for glibc (see feature_test_macros(7)):
 
       timer_create(): _POSIX_C_SOURCE >= 199309L
 
DESCRIPTION
       timer_create() creates a new per-process interval timer.  The ID of the new timer is returned in the buffer pointed to by timerid,  which  must be a non-null pointer.  This ID is unique within the process, until the  timer is deleted.  The new timer is initially disarmed.
       The clockid argument specifies the clock that the  new  timer  uses  to measure time.  It can be specified as one of the following values:
       CLOCK_REALTIME
              A settable system-wide real-time clock.
       CLOCK_MONOTONIC
              A  nonsettable monotonically increasing clock that measures time from some unspecified point in the past  that  does  not  change after system startup.
       CLOCK_PROCESS_CPUTIME_ID (since Linux 2.6.12)
              A  clock  that  measures  (user and system) CPU time consumed by (all of the threads in) the calling process.
       CLOCK_THREAD_CPUTIME_ID (since Linux 2.6.12)
              A clock that measures (user and system) CPU time consumed by the calling thread.
       As  well  as  the above values, clockid can be specified as the clockid returned  by  a  call  to  clock_getcpuclockid(3)  or   pthread_getcpuclockid(3).
       The sevp argument points to a sigevent structure that specifies how the  caller should be notified when the timer expires.  For  the  definition  and general details of this structure, see sigevent(7).
       The sevp.sigev_notify field can have the following values:
       SIGEV_NONE
              Don't asynchronously notify when the timer expires.  Progress of the timer can be monitored using timer_gettime(2).
       SIGEV_SIGNAL
              Upon timer expiration, generate the signal sigev_signo  for  the process.   See  sigevent(7)  for  general  details.  The si_code  field of the siginfo_t structure will be set  to  SI_TIMER.   At  any  point  in time, at most one signal is queued to the process  for a given timer; see timer_getoverrun(2) for more details.
       SIGEV_THREAD
              Upon timer expiration, invoke  sigev_notify_function  as  if  it were  the  start  function of a new thread.  See sigevent(7) for  details.
       SIGEV_THREAD_ID (Linux-specific)
              As for SIGEV_SIGNAL, but the signal is targeted  at  the  thread whose  ID  is  given  in sigev_notify_thread_id, which must be a  thread   in   the   same   process   as   the    caller.     The sigev_notify_thread_id  field specifies a kernel thread ID, that is, the value returned by clone(2) or gettid(2).  This  flag  is  intended only for use by threading libraries.
       Specifying  sevp  as  NULL  is  equivalent to specifying a pointer to a sigevent structure in which sigev_notify is  SIGEV_SIGNAL,  sigev_signo  is SIGALRM, and sigev_value.sival_int is the timer ID.
RETURN VALUE
       On  success,  timer_create()  returns 0, and the ID of the new timer is  placed in *timerid.  On failure, -1 is returned, and errno  is  set  to  indicate the error.
ERRORS
       EAGAIN Temporary error during kernel allocation of timer structures.
       EINVAL Clock  ID,  sigev_notify, sigev_signo, or sigev_notify_thread_id  is invalid.
       ENOMEM Could not allocate memory.
VERSIONS
       This system call is available since Linux 2.6.
CONFORMING TO
       POSIX.1-2001, POSIX.1-2008.
NOTES
       A program may create multiple interval timers using timer_create().
       Timers are not inherited by the child of a fork(2),  and  are  disarmed  and deleted during an execve(2).
       The kernel preallocates a "queued real-time signal" for each timer created using timer_create().  Consequently, the number of timers is  limited by the RLIMIT_SIGPENDING resource limit (see setrlimit(2)).
       The  timers  created  by  timer_create()  are  commonly known as "POSIX (interval) timers".  The POSIX timers API  consists  of  the  following interfaces:
       *  timer_create(): Create a timer.
       *  timer_settime(2): Arm (start) or disarm (stop) a timer.
       *  timer_gettime(2): Fetch the time remaining until the next expiration  of a timer, along with the interval setting of the timer.
       *  timer_getoverrun(2): Return the overrun count  for  the  last  timer  expiration.
       *  timer_delete(2): Disarm and delete a timer.
       Since  Linux  3.10, the /proc/[pid]/timers file can be used to list the POSIX timers for the process with PID pid.   See  proc(5)  for  further information.
   C library/kernel differences
       Part  of  the  implementation  of  the  POSIX timers API is provided by glibc.  In particular:
       *  Much of the functionality for  SIGEV_THREAD  is  implemented  within glibc,  rather  than the kernel.  (This is necessarily so, since the thread involved in handling the notification is  one  that  must  be  managed  by  the  C library POSIX threads implementation.)  Although  the notification delivered to the process is via  a  thread,  internally   the   NPTL  implementation  uses  a  sigev_notify  value  of SIGEV_THREAD_ID along with a real-time signal that  is  reserved  by  the implementation (see nptl(7)).
       *  The  implementation of the default case where evp is NULL is handled  inside glibc, which invokes the underlying system call with a  suitably populated sigevent structure.
       *  The timer IDs presented at user level are maintained by glibc, which maps these IDs to the timer IDs employed by the kernel.
       The POSIX timers system calls first appeared in Linux  2.6.   Prior  to  this,   glibc   provided   an   incomplete   user-space  implementation  (CLOCK_REALTIME timers only) using POSIX threads, and in glibc versions before 2.17, the implementation falls back to this technique on systems  running pre-2.6 Linux kernels.
EXAMPLE
       The program below takes two arguments: a sleep period in seconds, and a  timer  frequency in nanoseconds.  The program establishes a handler for  the signal it uses for the timer, blocks that signal, creates and  arms a timer that expires with the given frequency, sleeps for the specified  number of seconds, and then unblocks the timer signal.   Assuming  that the  timer  expired  at  least once while the program slept, the signal handler will be invoked, and  the  handler  displays  some  information  about the timer notification.  The program terminates after one invocation of the signal handler. In the following example run, the program sleeps for  1  second,  after creating  a timer that has a frequency of 100 nanoseconds.  By the time the signal is unblocked and delivered, there have been around ten  million overruns.
           $ ./a.out 1 100
           Establishing handler for signal 34
           Blocking signal 34
           timer ID is 0x804c008
           Sleeping for 1 seconds
           Unblocking signal 34
           Caught signal 34
               sival_ptr = 0xbfb174f4;     *sival_ptr = 0x804c008
               overrun count = 10004886
 
   Program source
 
       #include <stdlib.h>
       #include <unistd.h>
       #include <stdio.h>
       #include <signal.h>
       #include <time.h>
 
       #define CLOCKID CLOCK_REALTIME
       #define SIG SIGRTMIN
 
       #define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \
                               } while (0)
 
       static void
       print_siginfo(siginfo_t *si)
       {
           timer_t *tidp;
           int or;
 
           tidp = si->si_value.sival_ptr;
 
           printf("    sival_ptr = %p; ", si->si_value.sival_ptr);
           printf("    *sival_ptr = 0x%lx\n", (long) *tidp);
 
           or = timer_getoverrun(*tidp);
           if (or == -1)
               errExit("timer_getoverrun");
 
           else
               printf("    overrun count = %d\n", or);
       }
 
       static void
       handler(int sig, siginfo_t *si, void *uc)
       {
           /* Note: calling printf() from a signal handler is not
              strictly correct, since printf() is not async-signal-safe;
              see signal(7) */
 
           printf("Caught signal %d\n", sig);
           print_siginfo(si);
           signal(sig, SIG_IGN);
       }
 
       int
       main(int argc, char *argv[])
       {
           timer_t timerid;
           struct sigevent sev;
           struct itimerspec its;
           long long freq_nanosecs;
           sigset_t mask;
           struct sigaction sa;
 
           if (argc != 3) {
               fprintf(stderr, "Usage: %s <sleep-secs> <freq-nanosecs>\n",
                       argv[0]);
               exit(EXIT_FAILURE);
           }
 
           /* Establish handler for timer signal */
 
           printf("Establishing handler for signal %d\n", SIG);
           sa.sa_flags = SA_SIGINFO;
           sa.sa_sigaction = handler;
           sigemptyset(&sa.sa_mask);
           if (sigaction(SIG, &sa, NULL) == -1)
               errExit("sigaction");
 
           /* Block timer signal temporarily */
 
           printf("Blocking signal %d\n", SIG);
           sigemptyset(&mask);
           sigaddset(&mask, SIG);
           if (sigprocmask(SIG_SETMASK, &mask, NULL) == -1)
               errExit("sigprocmask");
 
           /* Create the timer */
 
           sev.sigev_notify = SIGEV_SIGNAL;
           sev.sigev_signo = SIG;
           sev.sigev_value.sival_ptr = &timerid;
 
           if (timer_create(CLOCKID, &sev, &timerid) == -1)
               errExit("timer_create");
 
           printf("timer ID is 0x%lx\n", (long) timerid);
 
           /* Start the timer */
 
           freq_nanosecs = atoll(argv[2]);
           its.it_value.tv_sec = freq_nanosecs / 1000000000;
           its.it_value.tv_nsec = freq_nanosecs % 1000000000;
           its.it_interval.tv_sec = its.it_value.tv_sec;
           its.it_interval.tv_nsec = its.it_value.tv_nsec;
 
           if (timer_settime(timerid, 0, &its, NULL) == -1)
                errExit("timer_settime");
 
           /* Sleep for a while; meanwhile, the timer may expire
              multiple times */
 
           printf("Sleeping for %d seconds\n", atoi(argv[1]));
           sleep(atoi(argv[1]));
 
           /* Unlock the timer signal, so that timer notification
              can be delivered */
 
           printf("Unblocking signal %d\n", SIG);
           if (sigprocmask(SIG_UNBLOCK, &mask, NULL) == -1)
               errExit("sigprocmask");
 
           exit(EXIT_SUCCESS);
       }
 
SEE ALSO
       clock_gettime(2), setitimer(2), timer_delete(2), timer_getoverrun(2),
       timer_settime(2), timerfd_create(2), clock_getcpuclockid(3),
       pthread_getcpuclockid(3), pthreads(7), sigevent(7), signal(7), time(7)
COLOPHON
       This page is part of release 4.04 of the Linux man-pages project.  A
       description of the project, information about reporting bugs, and the
       latest version of this page, can be found at
       http://www.kernel.org/doc/man-pages/.