lkml.org 
[lkml]   [2019]   [Jun]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH v5 01/18] kunit: test: add KUnit test runner core
Date
Quoting Brendan Higgins (2019-06-25 13:28:25)
> On Wed, Jun 19, 2019 at 5:15 PM Stephen Boyd <sboyd@kernel.org> wrote:
> >
> > Quoting Brendan Higgins (2019-06-17 01:25:56)
> > > diff --git a/kunit/test.c b/kunit/test.c
> > > new file mode 100644
> > > index 0000000000000..d05d254f1521f
> > > --- /dev/null
> > > +++ b/kunit/test.c
> > > @@ -0,0 +1,210 @@
> > > +// SPDX-License-Identifier: GPL-2.0
> > > +/*
> > > + * Base unit test (KUnit) API.
> > > + *
> > > + * Copyright (C) 2019, Google LLC.
> > > + * Author: Brendan Higgins <brendanhiggins@google.com>
> > > + */
> > > +
> > > +#include <linux/sched/debug.h>
> > > +#include <kunit/test.h>
> > > +
> > > +static bool kunit_get_success(struct kunit *test)
> > > +{
> > > + unsigned long flags;
> > > + bool success;
> > > +
> > > + spin_lock_irqsave(&test->lock, flags);
> > > + success = test->success;
> > > + spin_unlock_irqrestore(&test->lock, flags);
> >
> > I still don't understand the locking scheme in this code. Is the
> > intention to make getter and setter APIs that are "safe" by adding in a
> > spinlock that is held around getting and setting various members in the
> > kunit structure?
>
> Yes, your understanding is correct. It is possible for a user to write
> a test such that certain elements may be updated in different threads;
> this would most likely happen in the case where someone wants to make
> an assertion or an expectation in a thread created by a piece of code
> under test. Although this should generally be avoided, it is possible,
> and there are occasionally good reasons to do so, so it is
> functionality that we should support.
>
> Do you think I should add a comment to this effect?

No, I think the locking should be removed.

>
> > In what situation is there more than one thread reading or writing the
> > kunit struct? Isn't it only a single process that is going to be
>
> As I said above, it is possible that the code under test may spawn a
> new thread that may make an expectation or an assertion. It is not a
> super common use case, but it is possible.

Sure, sounds super possible and OK.

>
> > operating on this structure? And why do we need to disable irqs? Are we
> > expecting to be modifying the unit tests from irq contexts?
>
> There are instances where someone may want to test a driver which has
> an interrupt handler in it. I actually have (not the greatest) example
> here. Now in these cases, I expect someone to use a mock irqchip or
> some other fake mechanism to trigger the interrupt handler and not
> actual hardware; technically speaking in this case, it is not going to
> be accessed from a "real" irq context; however, the code under test
> should think that it is in an irq context; given that, I figured it is
> best to just treat it as a real irq context. Does that make sense?

Can you please describe the scenario in which grabbing the lock here,
updating a single variable, and then releasing the lock right after
does anything useful vs. not having the lock? I'm looking for a two CPU
scenario like below, but where it is a problem. There could be three
CPUs, or even one CPU and three threads if you want to describe the
extra thread scenario.

Here's my scenario where it isn't needed:

CPU0 CPU1
---- ----
kunit_run_test(&test)
test_case_func()
....
[mock hardirq]
kunit_set_success(&test)
[hardirq ends]
...
complete(&test_done)
wait_for_completion(&test_done)
kunit_get_success(&test)

We don't need to care about having locking here because success or
failure only happens in one place and it's synchronized with the
completion.

>
> > > +
> > > + return success;
> > > +}
> > > +
> > > +static void kunit_set_success(struct kunit *test, bool success)
> > > +{
> > > + unsigned long flags;
> > > +
> > > + spin_lock_irqsave(&test->lock, flags);
> > > + test->success = success;
> > > + spin_unlock_irqrestore(&test->lock, flags);
> > > +}

\
 
 \ /
  Last update: 2019-06-26 05:42    [W:0.092 / U:58.144 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site