LTP - gettimeofday02 FAIL

I've been getting a somewhat random error in a few of the recent 2.5 kernels with SMP machines. I noticed this on a 2.5.47 bk pull, but I was also able to reproduce it on 2.5.46. I haven't tried any earlier kernels yet. The LTP gettimeofday02 test sometimes fails with this message:
gettimeofday02 0 INFO : checking if gettimeofday is monotonous,
takes 30s
gettimeofday02 1 FAIL : Time is going backwards (old
1037138184.846333 vs new 1037138184.843346!

I have not been able to reproduce this on a single processor machine though.

Basically, all the test does is:
gettimeofday(&tv1, NULL);

while(!done) {
	gettimeofday(&tv2, NULL);
	FAIL if tv2 < tv1
	tv1 = tv2;
}

Any ideas on what could be causing this?

Thanks,
Paul Larson

On Tue, Nov 12, 2002 at 04:11:14PM -0600, Paul Larson wrote:

    I have not been able to reproduce this on a single processor machine
    though.

    Basically, all the test does is:
    gettimeofday(&tv1, NULL);
    while(!done) {
    	gettimeofday(&tv2, NULL);
    	FAIL if tv2 < tv1
    	tv1 = tv2;
    }

    Any ideas on what could be causing this?

The TSC's aren't synchronized between CPUs.

This is becoming more and more of a problem and in-escapable on some hardware so I'm starting to wonder if assuming the TSCs are even roughly synchronized *anywhere* is a good idea.

--cw

So this is a hardware issue? no way around this?

-Paul Larson

For some people it is a hardware issue. The way around is not not use the TSC for certain things.

--cw