RTEMS Tester

Chris Johns <chrisj@rtems.org> 1.0, June 2014

1. RTEMS Tester

The RTEMS Tester is a test framework. It includes a command line interface to run tests on supported targets. The framework provides backend support for common simulators and debuggers. The board support package (BSP) configurations for RTEMS are provided and can be used to run all the tests provided with RTEMS. The framework is not specific to RTEMS and can be configured to run any suitable application.

RTEMS is an embedded operating system and is cross-compiled on a range of host machines. The executables run on the target hardware and this can vary widely from open source simulators, commercial simulators, debuggers with simulators, to debuggers with hardware specific pods and devices. Testing RTEMS requires the cross-compiled test executable is transfered to the target hardware, executed and the output returned to the host where it is analyised to determine the test result. The RTEMS Tester provides a framework to do this.

Running all the RTEMS tests on your target is very important. It provides you with a tracable record your RTEMS version and its tools and working at the level the RTEMS development team expect when releasing RTEMS. Being able to easly run the tests and verify the results is critical in maintiaining a high standard.

The RTEMS Tester contains:

Command line tool (rtems-test)
BSP Configuration scripts
Backend Configuration scripts
Backend Python classes
Python based framework

If you have a problem please see the reporting bugs section.

This document is for users of the RTEMS Tester as well as those wanting to develop, extend and investigate the RTEMS Tester.

2. License

The RTEMS Tester is part of the RTEMS Tools Project. The code is released under the OSI approved The BSD 2-Clause License. It is free to use and we encourage this including operating systems other than RTEMS.

The code and command line tools must retain the same names and always reference the RTEMS Tools Project.

3. Quick Start

The quick start will show you how to run the test suite for a BSP. It will explain how to get the RTEMS Tester, set it up and run the tests for the SIS BSP. It assumes you have a valid SPARC tool chain and built SIS BSP version of RTEMS. 4.11.

3.1. Setup

Setup a development work space:

$ cd
$ mkdir -p development/rtems/test
$ cd development/rtems/test

First fetch the RTEMS tester from the RTEMS Tools repository

$ git git://git.rtems.org/rtems-tools.git rtems-tools.git
$ cd rtems-tools.git/tester

3.2. Available BSPs

You can list the available BSP’s with:

$ ./rtems-test --list-bsps
  arm920
  beagleboardxm
  jmr3904-run
  jmr3904
  mcf5235
  psim-run
  psim
  realview_pbx_a9_qemu
  sis-run
  sis
  xilinx_zynq_a9_qemu
  xilinx_zynq_a9_qemu_smp
  xilinx_zynq_zc706
  xilinx_zynq_zc706_qemu

The list is growing all the time and if your BSP is not supported we encourage you to add it and submit the configuration back to the project.

Some of the BSPs may appear more than once in the list. These are aliased BSP configuration’s that may use a different backend. An example is the SPARC Instruction Simulator (SIS) BSP. There is the sis BSP which uses the GDB backend and the sis-run which uses the command line version of the SIS simulator. We will show how to use rtems-test conmand with the SIS BSP because it is easy to build an use.

3.3. Building RTEMS Tests

Build RTEMS with a configuration command line something similar to:

The following assumes a Unix type host and the tools have been built with a prefix of $HOME/development/rtems/4.11.

$ cd
$ export PATH=$HOME/development/rtems/4.11/bin:$PATH
$ mkdir -p development/rtems/kernel
$ cd development/rtems/kernel
$ git clone git://git.rtems.org/rtems.git rtems.git
$ cd rtems.git
$ ./bootstrap -c && ./bootstrap -p && ./bootstrap
$ cd ..
$ mkdir -p builds/sis
$ cd builds/sis
$ ../../rtems.git/configure --target=sparc-rtems4.11 \
                    --enable-tests --enable-rtemsbsp=sis
$ make

Add the -j option with the number of cores to run the build parallel where it can.

Building all the tests takes time and it uses more disk so be patient. When finished all the tests will be built ready to run. Before running all the tests it is a good idea to run the hello test. The hello test is an RTEMS version of the classic "Hello World" example and running it shows you have a working tool chain and build of RTEMS ready to run the tests. Using the run command:

$ sparc-rtems4.11-run  sparc-rtems4.11/c/sis/testsuites/samples/hello/hello.exe

*** BEGIN OF TEST HELLO WORLD ***
Hello World
*** END OF TEST HELLO WORLD ***

The run command is the GDB simulator without the GDB part.

Running the example using GDB:

$ sparc-rtems4.11-gdb sparc-rtems4.11/c/sis/testsuites/samples/hello/hello.exe
GNU gdb (GDB) 7.7
Copyright (C) 2014 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "--host=x86_64-apple-darwin13.2.0
--target=sparc-rtems4.11".
Type "show configuration" for configuration details.
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
Find the GDB manual and other documentation resources online at:
<http://www.gnu.org/software/gdb/documentation/>.
For help, type "help".
Type "apropos word" to search for commands related to "word"...
Reading symbols from
sparc-rtems4.11/c/sis/testsuites/samples/hello/hello.exe...done.
(gdb) target sim
Connected to the simulator.
(gdb) load
(gdb) r 
Starting program: sparc-rtems4.11/c/sis/testsuites/samples/hello/hello.exe


*** BEGIN OF TEST HELLO WORLD ***
Hello World
*** END OF TEST HELLO WORLD ***
[Inferior 1 (process 42000) exited normally]
(gdb) q

To debug set break points before issuing the GDB run command.

There are currently close to 500 separate tests and you can run them all with a single RTEMS Tester command.

Use a recent RSB version to build a SPARC tool chain. A recent patch fixes issues with the GDB simulator and it improves the test results. It is also recommended you use a recent RTEMS 4.11 version as the tests have been cleaned up to improve the results.

3.4. Running the Tests

The rtems-test command line accepts a range of options. These are discussed later in the manual. Any command line argument without a -- prefix is a test executable. You can pass more than one executable on the command line. If the executable is a path to a directory the directories under that path are searched for any file with a .exe extension. This is the detault extension for RTEMS executables built within RTEMS.

To run the SIS tests enter the following command from the top of the SIS BSP build tree:

$ ~/development/rtems/test/rtems-tools.git/tester/rtems-test  \
             --log=log_sis_run  \
             --rtems-bsp=sis-run  \
             --rtems-tools=$HOME/development/rtems/4.11  \
                 sparc-rtems4.11/c/sis/testsuites 
RTEMS Testing - Tester, v0.2.0
[  1/500] p:0   f:0   t:0   i:0   | sparc/sis: fsbdpart01.exe
[  2/500] p:0   f:0   t:0   i:0   | sparc/sis: fsdosfsformat01.exe
[  5/500] p:0   f:0   t:0   i:0   | sparc/sis: fsdosfswrite01.exe
[  4/500] p:0   f:0   t:0   i:0   | sparc/sis: fsdosfssync01.exe
[  7/500] p:0   f:0   t:0   i:0   | sparc/sis: fsimfsgeneric01.exe
[  3/500] p:0   f:0   t:0   i:0   | sparc/sis: fsdosfsname01.exe
[  8/500] p:0   f:0   t:0   i:0   | sparc/sis: fsnofs01.exe
[  6/500] p:0   f:0   t:0   i:0   | sparc/sis: fsfseeko01.exe
[  9/500] p:5   f:0   t:0   i:0   | sparc/sis: fsrfsbitmap01.exe
[ 10/500] p:5   f:0   t:0   i:0   | sparc/sis: imfs_fserror.exe
[ 12/500] p:5   f:0   t:0   i:0   | sparc/sis: imfs_fspatheval.exe
  ................................................. 
[492/500] p:482 f:0   t:2   i:0   | sparc/sis: tm25.exe
[493/500] p:483 f:0   t:2   i:0   | sparc/sis: tm26.exe
[495/500] p:485 f:0   t:2   i:0   | sparc/sis: tm28.exe
[494/500] p:485 f:0   t:2   i:0   | sparc/sis: tm27.exe
[496/500] p:487 f:0   t:2   i:0   | sparc/sis: tm29.exe
[497/500] p:487 f:0   t:2   i:0   | sparc/sis: tm30.exe
[498/500] p:489 f:0   t:2   i:0   | sparc/sis: tmck.exe
[499/500] p:490 f:0   t:2   i:0   | sparc/sis: tmcontext01.exe
[500/500] p:490 f:0   t:2   i:0   | sparc/sis: tmoverhd.exe
Passed:   495 
Failed:     0
Timeouts:   5
Invalid:    0
Total:    500
Average test time: 0:00:00.608249 
Testing time     : 0:05:04.124696

	The RTEMS Tester’s test command. In this example we are using an absolute path.
	The --log option sends the output to a log file. By default only failed tests log the complete output.
	Select the SIS BSP and use GDB.
	Path to the RTEMS tools so GDB can be found.
	Path to the SIS BSP built with all tests to run. If you add subdirectories to the path specific tests can be run.
	The output has been shortened so it fits nicely here.
	The test results. It shows passes, fails, timeouts, and invalid results. In this run 495 tests passed and 5 tests timedout. The timeouts are probability due the tests not having enough execute time to complete. The default timeout is 180 seconds and some of the interrupt tests need longer. The amount of time depends on the performance of your host CPU running the simulations.
	The average time per test and the total time taken to run all the tests.

This BSP requires the --rtems-tools option because the SPARC GDB is the sparc-rtems4.11-gdb command that is part of the RTEMS tools. Not every BSP will require this option so you will need to check the specifics of the BSP configration to determine if it is needed.

The output you see is each test starting to run. The rtems-test command can run multiple SIS GDB simulations in parallel so you will see a number start quickly and then tests start as others finish. The output shown here is from an 8 core processor so the first 8 are started in parallel and the status shows the order they actually started which is not 1 to 8.

The test start line shows the current status of the tests. The status reported is when the test starts and not the result of that test. A fail, timeout or invalid count changing means a test running before this test started failed, not the starting test. The status here has 495 tests pass and no failures and 5 timeouts.:

[295/500] p:287 f:0  t:1  i:0 | sparc/sis: sp26.exe

	The test number, in this case test 295 of 500 tests.
	Passed test count.
	Failied test count.
	Timeout test count.
	Invalid test count.
	Architecture and BSP.
	Executable name.

The test log records all the tests and results. The reporting mode by default only provides the output history if a test fails, timeouts, or is invalid. The time taken by each test is also recorded.

The tests must complete in a specified time or the test is marked as timed out. The default timeout is 3 minutes and can be globally changed using the --timeout command line option. The time required to complete a test can vary. When simulators are run in parallel the time taken depends on the specifics of the host machine being used. A test per core is the most stable method even though more tests can be run than available cores. If your machine needs longer or you are using a VM you may need to lengthen the time out.

3.5. Test Status

Tests can be marked with one of the following:

Pass
Fail
Timeout
Invalid

The RTEMS console or stdout output from the test is needed to determine the result of the test.

Pass

A test passes if the start and end markers are seen in the test output. The start marker is "***" and the end mark is "*** END OF TEST". All tests in the RTEMS test suite have these markers.

Fail

A test fails if the start marker is seen and there is no end marker.

Timeout

If the test does not complete within the timeout setting the test is marked as timed out.

Invalid

If no start marker is seen the test is marked as invalid. If you are testing on real target hardware things can sometimes go wrong and the target may not initialise or respond to the debugger in an expected way.

3.6. Reporting

The report written to the log has the following modes:

All (all)
Failures (failures)
None (none)

The mode is controlled using the command line option --report-mode using the values listed above.

All

The output of all tests is written to the log.

Failures

The output of the all tests that do not pass is written to the log.

None

No output is written to the log.

The output is tagged so you can determine where it comes from. The following is the complete output for the In Memory File System test imfs_fslink.exe running on a Coldfire MCF5235 using GDB and a BDM pod:

[ 11/472] p:9   f:0   t:0   i:1   | m68k/mcf5235: imfs_fslink.exe
> gdb: ..../bin/m68k-rtems4.11-gdb -i=mi --nx --quiet ..../imfs_fslink.exe 
> Reading symbols from ..../fstests/imfs_fslink/imfs_fslink.exe...
> done. 
> target remote | m68k-bdm-gdbserver pipe 003-005
> Remote debugging using | m68k-bdm-gdbserver pipe 003-005
> m68k-bdm: debug module version 0
> m68k-bdm: detected MCF5235
> m68k-bdm: architecture CF5235 connected to 003-005
> m68k-bdm: Coldfire debug module version is 0 (5206(e)/5235/5272/5282)
> Process 003-005 created; pid = 0
> 0x00006200 in ?? ()
> thb *0xffe254c0
> Hardware assisted breakpoint 1 at 0xffe254c0
> continue
> Continuing.
] 
]
] External Reset
]
] ColdFire MCF5235 on the BCC
] Firmware v3b.1a.1a (Built on Jul 21 2004 17:31:28)
] Copyright 1995-2004 Freescale Semiconductor, Inc.  All Rights Reserved.
]
] Enter 'help' for help.
]
> Temporary breakpoint
> 1, 0xffe254c0 in ?? ()
> load
> Loading section .text, size 0x147e0 lma 0x40000
> Loading section .data, size 0x5d0 lma 0x547e0
> Start address 0x40414, load size 85424
> Transfer rate: 10 KB/sec, 1898 bytes/write.
> b bsp_reset
> Breakpoint 2 at 0x41274: file ..../shared/bspreset_loop.c, line 14.
> continue
> Continuing.
] dBUG>
]
] *** FILE SYSTEM TEST ( IMFS ) *** 
] Initializing filesystem IMFS
]
]
] *** LINK TEST ***
] link creates hardlinks
] test if the stat is the same
] chmod and chown
] unlink then stat the file
] *** END OF LINK TEST ***
]
]
] Shutting down filesystem IMFS
] *** END OF FILE SYSTEM TEST ( IMFS ) *** 
> Breakpoint
> 2, bsp_reset () at ..../m68k/mcf5235/../../shared/bspreset_loop.c:14
> 14    {
Result: passed     Time: 0:00:10.045447

	GDB command line (Note: paths with '….' have been shortened)
	Lines starting with > are from GDB’s console.
	Line starting with ] are from the target’s console.
	The start marker.
	The end marker.
	The result with the test time.

4. Running Tests in Parallel

The RTEMS Tester supports parallel execution of tests by default. This only makes sense if the test backend can run in parallel without resulting in resource contention. Simulators are an example of backends that can run in parallel. A hardware debug tool like a BDM or JTAG pod can only a single test at once to the tests need to be run one at a time.

The test framework manages the test jobs and orders the output in the report log in test order. Output is held for completed tests until the next test to be reported has finished.

4.1. Command Line Help

The rtems-test command line accepts a range of options. You can review the available option by the --help option:

RTEMS Tools Project (c) 2012-2014 Chris Johns
Options and arguments:
--always-clean               : Always clean the build tree, even with an error
--debug-trace                : Debug trace based on specific flags
--dry-run                    : Do everything but actually run the build
--force                      : Force the build to proceed
--jobs=[0..n,none,half,full] : Run with specified number of jobs, default: num CPUs.
--keep-going                 : Do not stop on an error.
--list-bsps                  : List the supported BSPs
--log file                   : Log file where all build out is written too
--macros file[,file]         : Macro format files to load after the defaults
--no-clean                   : Do not clean up the build tree
--quiet                      : Quiet output (not used)
--report-mode                : Reporting modes, failures (default),all,none
--rtems-bsp                  : The RTEMS BSP to run the test on
--rtems-tools                : The path to the RTEMS tools
--target                     : Set the target triplet
--timeout                    : Set the test timeout in seconds (default 180 seconds)
--trace                      : Trace the execution
--warn-all                   : Generate warnings

5. Developement

The RTEMS Tester framework and command line tool is under active development. This are changing, being fixed, broken and generally improved. If you want to help please see the Wiki page for open itmes.

6. History

The RTEMS Tester is based on a refactored base of Python code used in the RTEMS Source Builder. This code provided a working tested base that has been extended and expanded to meet the needs of the RTEMS Tester. The tester uses the specifics found in the various scripts and configurations in the rtems-testing.git repo that has been accumulated over many years. The shell script implementation is restricted in what can it do and per BSP script is a maintenance burden, for example the command lines and options vary between each script.