Profiling timer not delivered during syscalls?

[Patrick J. LoPresti]

I did this test on a Red Hat 6.1 system (glibc 2.1.2) with kernel
2.2.13.

I compiled the appended program like so:

   gcc -O2 -Wall -g -pg -o foo foo.c

*** Note the -pg option; it is the whole point of the exercise ***

I ran it, and found that the profiling timer is not being delivered
while the program is in select().  In fact, I have not been able to
find a single system call during which the profiling signal is
delivered.  (I do admit, so far I have only tried select(), sleep(),
and pthread_join().)

Is this behavior expected?  I would have thought that time spent in
system calls would be pretty important when profiling an application.

 - Pat

======================================================================
#include <stdio.h>
#include <signal.h>
#include <unistd.h>
#include <sys/time.h>
#include <sys/types.h>

static __sighandler_t orig_sigprof_handler;

static void
handle_sigprof (int sig, struct sigcontext cxt)
{
  fprintf (stderr, "here\n");

  if (orig_sigprof_handler != NULL)
    ((void (*)(int, struct sigcontext))orig_sigprof_handler) (sig, cxt);
}

static void
do_nothing (void)
{
  return;
}

int
main (int argc, char *argv[])
{
  struct sigaction act;
  int i;
  struct timeval timeout = { 10, 0 };

  /* Get old signal handler. */
  if (__sigaction (SIGPROF, NULL, &act) != 0)
    {
      perror ("first sigaction");
      exit (1);
    }

  /* Install new signal handler, preserving flags and such. */
  orig_sigprof_handler = act.sa_handler;
  act.sa_handler = (__sighandler_t)handle_sigprof;
  if (__sigaction (SIGPROF, &act, NULL) != 0)
    {
      perror ("second sigaction");
      exit (1);
    }

  select (0, NULL, NULL, NULL, &timeout);
  fprintf (stderr, "select finished\n");
  for (i=0 ; i<2000000 ; i++)
    do_nothing();

  return 0;
}