Copyright (c) 2015 Netflix, Inc.
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
1. Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
2. Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
THIS SOFTWARE IS PROVIDED BY THE DEVELOPERS ``AS IS'' AND ANY EXPRESS OR
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
IN NO EVENT SHALL THE DEVELOPERS BE LIABLE FOR ANY DIRECT, INDIRECT,
INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
$FreeBSD$
.Dd November 12, 2015 .Dt KERN_TESTFRWK 9 .Os .Sh NAME .Nm kern_testfrwk .Nd A kernel testing framework .Sh SYNOPSIS kldload kern_testfrwk .Sh DESCRIPTION This whole section is not written in manual page style and should be ripped
out and replaced. -CEM
So what is this sys/tests directory in the kernel all about?
p Have you ever wanted to test a part of the .Fx kernel in some way and you had no real way from user-land to make what you want to occur happen? Say an error path or situation where locking occurs in a particular manner that happens only once in a blue moon?
p If so, then the kernel test framework is just what you are looking for. It is designed to help you create the situation you want.
p There are two components to the system: the test framework and your test. This document will describe both components and use the test submitted with the initial commit of this code to discuss the test .Xr ( callout_test 4 ) . All of the tests become kernel loadable modules. The test you write should have a dependency on the test framework. That way it will be loaded automatically with your test. For example, you can see how to do this in the bottom of callout_test.c in
a sys/tests/callout_test/callout_test.c .
p The framework itself is in
a sys/tests/framework/kern_testfrwk.c . Its job is to manage the tests that are loaded. (More than one can be loaded.) The idea is pretty simple; you load the test framework and then load your test.
p When your test loads, you register your tests with the kernel test framework. You do that through a call to .Fn kern_testframework_register . Usually this is done at the module load event as shown below: d -literal -offset indent switch (type) { case MOD_LOAD: err = kern_testframework_register("callout_test", run_callout_test); .Ed
p Here the test is "callout_test" and it is registered to run the function .Fn run_callout_test passing it a .Fa struct kern_test *ptr . The .Vt kern_test structure is defined in
a kern_testfrwk.h . d -literal -offset indent struct kern_test { char name[TEST_NAME_LEN]; int num_threads; /* Fill in how many threads you want */ int tot_threads_running; /* Private to framework */ uint8_t test_options[TEST_OPTION_SPACE]; }; .Ed
p The user sends this structure down via a sysctl to start your test. He or she places the same name you registered ("callout_test" in our example) in the .Va name field. The user can also set the number of threads to run with .Va num_threads .
p The framework will start the requested number of kernel threads, all running your test at the same time. The user does not specify anything in .Va tot_threads_running ; it is private to the framework. As the framework calls each of your tests, it will set the .Va tot_threads_running to the index of the thread that your call is made from. For example, if the user sets .Va num_threads to 2, then the function .Fn run_callout_test will be called once with .Va tot_threads_running to 0, and a second time with .Va tot_threads_running set to 1.
p The .Va test_options field is a test-specific set of information that is an opaque blob. It is passed in from user space and has a maximum size of 256 bytes. You can pass arbitrary test input in the space. In the case of callout_test we reshape that to: d -literal -offset indent struct callout_test { int number_of_callouts; int test_number; }; .Ed
p
So the first lines of
.Fn run_callout_test
does the following to get at the user specific data:
This is a bad example and violates strict aliasing. It should be replaced.
d -literal -offset indent struct callout_test *u;
size_t sz;
int i;
struct callout_run *rn;
int index = test->tot_threads_running;
u = (struct callout_test *)test->test_options;
.Ed
p That way it can access: .Va u->test_number (there are two types of tests provided with this test) and .Va u->number_of_callouts (how many simultaneous callouts to run).
p Your test can do anything with these bytes. So the callout_test in question wants to create a situation where multiple callouts are all run, that is the .Va number_of_callouts , and it tries to cancel the callout with the new .Fn callout_async_drain . The threads do this by acquiring the lock in question, and then starting each of the callouts. It waits for the callouts to all go off (the executor spins waits). This forces the situation that the callouts have expired and are all waiting on the lock that the executor holds. After the callouts are all blocked, the executor calls .Fn callout_async_drain on each callout and releases the lock.
p
callout_test(4) specific documentation should probably be moved to its own
page.
After all the callouts are done, a total status is printed
showing the results via
.Xr printf 9 .
The human tester can run
.Xr dmesg 8
to see the results.
In this case it is expected that if you are running test 0, all the callouts
expire on the same CPU so only one callout_drain function would have been
called.
the number of zero_returns should match the number of callout_drains that were
called, i.e., 1.
The one_returns should be the remainder of the callouts.
If the test number was 1, the callouts were spread across all CPUs.
The number of zero_returns will again match the number of drain calls made
which matches the number of CPUs that were put in use.
p More than one thread can be used with this test, though in the example case it is probably not necessary.
p You should not need to change the framework. Just add tests and register them after loading. .Sh AUTHORS The kernel test framework was written by .An Randall Stewart Aq Mt rrs@FreeBSD.org with help from .An John Mark Gurney Aq Mt jmg@FreeBSD.org .