xref: /external/ltp/testcases/open_posix_testsuite/functional/threads/pi_test

Real Time NPTL(rt-nptl) test suite README
=========================================

The document describes what rt-nptl tests are, how they are created,
how to make and run the tests and how to analysis the result.

rt-nptl tests include two parts, one is robust mutex tests, the other
is priority inheritance(pi) mutex tests.

In the document, <rtnptl-tests> stands for the directory where
rtnptl-tests was extracted.


Building the tests
-------------------
If you use the build method outlined in INSTALL-rtnptl, then just
change the top Makefile and the two Makefile under pi_test and
robust_test to include the Makefile.inc generated by INSTALL-rtnptl.

Otherwise, you need to set the CFLAGS and LDFLAGS of the Makefile to
point to where your copy of glibc+RTNPTL is built.

Run the tests
----------------
Use ./run.sh to run the tests; this makes sure there is a basic
watchdog timer to kill the pi mutex tests if they fail.

For different tests, there are different result analysis methods. The
following two sections will describe this in detail.

Priority Inheritance Mutex Tests
--------------------------
The tests are under <rtnptl-tests>/pi_test directory.

rt-nptl supports priority inheritance, if task TL with priority P(TL)
held a mutex, task TB with priority P(TB) claims the mutex and becomes
a waiter, P(TB)>P(TL), TL will boost to TB's priority until it
releases the lock. The pi mutex tests create several scenarios to test
this functionality. Please refer to the description in the tests for
the detailed information.

Currently pi mutex tests don't provide automatic PASS or FAIL result
display. Although users can use run.sh under <rtnptl-tests>/pi_test to
run the pi mutex tests, they need to analyze the output to get the
PASS or FAIL test result manually. After executing run.sh, the output
files will be generated with the name output.<testcasename>, such as
output.pitest-1.

If users use run.sh from ssh, to avoid the priority inversion
problem happened occasionally, users can use "chrt -p -f $PRIORITY
$$PPID" to increase the priority of sshd, if use run.sh from console,
users can use "chrt -p -f $PRIORITY $$" to increase the shell priority
before executing run.sh.

A sampling thread will sample the progress of working threads (TL, TPs
and TFs) at a specified interval and print out an integer value.
Progress of TL and TP is measured by how many times they execute a
busy loop; the output of this count indicates how far they progress
with respect to each other.

The first column of the output is the time when sampling happens. The
second column indicates the progress of TL thread at that time. The
third column indicates the progress of TP at that time. For pitest-2
and pitest-3, the second column is TP1 and the third column is TP2.
The remaining columns indicate the progress of TF (if any).

If the priority of TL is higher than (or equal to) the priority of TP,
TP will make not progress (the output of TP will not increase, since
it will not be scheduled to execute). Otherwise, TP and TL will both
make progress. As a result, Users can observe the priority change of
TL by the progress of TP.

Users can also use 'do-plot' utility under <rtnptl-tests>/pi_test to
generate a diagram using the output of test case. To do so:

1. Redirect the output of test case into a file, e.g:
	pitest-1 > output.pitest-1

2. Under X term, run:
	do-plot output.pitest-1

do-plot will invoke 'gnuplot' to generate a diagram showing the
progress of TL and TP. ('gnuplot has to be installed in the system').

Given pitest-1 as an example:
If the pi function works, TF's progress should be a straight line
constantly going up; TP's progress should start at about 10 seconds
and should be parallel to TF's until ten seconds later (20 seconds)
when TL is created; at this time TP's slope should go down a wee bit
and TL's slope should not be zero. After 10 seconds (30 seconds), TB
is created and TL boosted. At this moment, TP's slope should go down
to zero and TL's should be parallel to TF's. After TB timeouts on
waiting the mutex (50 seconds), TL and TP's slope will change back to
the original trend during 20~30 seconds.

You can refer to
http://developer.osdl.org/dev/robustmutexes/pitest-1-0.5.png for the
diagram generated by do-plot for pitest-1.