bin/49144: udf_times test takes 951 times longer on i386 than amd64

[gson%gson.org@localhost (Andreas Gustafsson)]

>Number:         49144
>Category:       bin
>Synopsis:       udf_times test takes 951 times longer on i386 than amd64
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    bin-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Aug 23 17:20:00 +0000 2014
>Originator:     Andreas Gustafsson
>Release:        NetBSD-current
>Organization:

>Environment:
System: NetBSD
Architecture: i386
Machine: i386
>Description:

This is a strange one...

The fs/vfs/t_unpriv/udf_times test case takes hundreds of times 
longer to run on NetBSD/i386 than it does on NetBSD/amd64.

Here is the atf-report output from running the test on NetBSD/amd64
and NetBSD/i386, both built from identical sources (-current CVS date
2014.08.17.20.22.41) and running on the same physical machine (an AMD
Athlon 64), on the bare metal:

  amd64:     udf_times: [0.086259s] Passed.
  i386:      udf_times: [82.081873s] Passed.

That's 951 times longer on i386 than amd64.

When running under qemu on the TNF testbed, both tests run a lot
slower as you would expect, and there, too, the i386 runs take
hundreds of times longer than the amd64 ones.  As a result, they
sometimes exceed the ATF timeout and fail.  On qemu/sparc, the tests
run about as fast as they do on qemu/amd64, so this does not look like
a 32-bit vs 64-bit issue.

According to the logs, the problem has existed at least since CVS date
2013.07.02.15.05.53, which was the first time the udf_times test ran
on babylon5 without running into other problems first.

>How-To-Repeat:

Run the ATF tests on i386.  Notice how the fs/vfs/t_unpriv/udf_times
test case takes minutes rather than milliseconds it takes on other
architectures.

>Fix: