I'd like to add this log message is not unique to my build and my Snow Leopard
environment.  Others have confirmed similar results.  See my post (xyzzy-xyzzy)
and the following posts in the the Apple Discussions:

https://discussions.apple.com/thread/6739506?start=30&tstart=0

A copy of a typical system.log file is also shown in that thread.  There are
other "mysterious" log entries as well (KOD?) but they are probably separate
issues from the one I am reporting here.

The only place in the code that we call mlockall() is line 800 of
ntpd/ntpd.c .

The call did not return the expected value for success, 0, so we
reported the error condition that we got back from the system.

Yes, I know where it was coming from (and I mention that further down in that
apple discussions thread).  But WHY is it happening?  And is it serious enough
to be concerned?

Note, while mlockall(MCL_CURRENT|MCL_FUTURE) is a valid call (I think) I cannot
understand why it should report an error on Snow Leopard (maybe other OS X's as
well, but I can't verify that).  Maybe there are some restrictions on when you
can use a mlockall(), some protocol needs to be followed, or perhaps under
certain conditions upon which ntpd is invoked as a process.  I don't know
enough about mlockall() to understand it.

I wrote a little test for mlockall():

#include <stdio.h>
#include <sys/mman.h>
#include <errno.h>
#include <string.h>

int main()
{
    int i, e;
    i = mlockall(MCL_CURRENT|MCL_FUTURE);
    e = errno;
    fprintf(stderr, "mlockall(MCL_CURRENT|MCL_FUTURE) returned %d (errno = %d
[%s])\n", i, e, strerror(e));
    return (i);
}

Here's what it prints:

mlockall(MCL_CURRENT|MCL_FUTURE) returned -1 (errno = 78 [Function not
implemented])

And it doesn't matter if the test is run in user space or under sudo.  So
mlockall() does indeed fail in Snow Leopard :-(  The question now is what can
be done about it?  As I asked earlier does it really matter that the mlockall()
doesn't lock ntpd's pages?  Why does configure think mlockall() is available
when apparently it isn't?  Should the ntp build be changed to handle such
cases?

Apparently, on that release of the OS the mlockall() function is
provided (so we detect it and believe we can call it) but the system
doesn't actually implement the appropriate functionality, so it returns
-1, with an errno of ENOSYS (function not implemented).

So basically this means that I have to "manually" change config.h after running
configure to redefine HAVE_MLOCKALL to 0 if I really want to suppress the log
messages.

I still think configure should be a little smarter when testing for mlockall()
to test its error return and define HAVE_MLOCKALL to 0 itself in config.h. 

Do I reason correctly that all the mlockall() was trying to do in ntpd was
prevent its process pages from being swapped out so that without it all that
results is a little less efficiency?  Thus the log messages could be safely
ignored without worry?

If somebody Knows a particular series of a given OS will NEVER support a
function like this we can suppress checking for it.  Otherwise, if it might
show up in a later version and the current binary might run there, then many
would argue that we want the call to be in there so an upgrade will just "do
the right thing".

We generally don't test to be sure things like this work, for two reasons.  One
is the one just given, and the other is we cannot run "executable" tests in
cross-compile environments, so the testing gets way uglier.  Three reasons. 
There are three ...  The third being it takes longer, and configure already
takes a very long time to execute.

I'm going to mark this as RESOLVED/WORKSFORME; if you disagree please feel free
to re-open this, perhaps after more discussion (perhaps on hackers@ or
questions@) so we can know how we should proceed.

I guess I can agree with you on the configure stuff.

But I still wish you would answer my questions I posed in my last update to
this thread.  I'll repeat them again here (slightly edited) for your
convenience...

Do I reason correctly that all the mlockall() was trying to do in ntpd was
prevent its process pages from being swapped out so that without it all that
results is a little less efficiency?  Thus the log messages could be safely
ignored without worry?  And its my choice if I want to suppress the log message
by me setting HAVE_MLOCKALL to 0 in config.h?

ntpd will mlockall() if available so that it can better synchronize the time on
your computer.

If you put:

 rlimit memlock 0

into your ntp.conf file that should also disable this.

Or you could #undef HAVE_MLOCKALL in config.h (#define'ing it to 0 probably
won't work).