Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220713005221.1926290-1-davidgow@google.com>
In-Reply-To: <20220713005221.1926290-1-davidgow@google.com>
From:   Daniel Latypov <dlatypov@google.com>
Date:   Wed, 13 Jul 2022 08:24:32 -0700
Message-ID: <CAGS_qxrNKnrWXhOfptz9iL5c_sixhKjpAfR2RLQi1XqL6m2Tpg@mail.gmail.com>
Subject: Re: [PATCH] module: kunit: Load .kunit_test_suites section when CONFIG_KUNIT=m
To:     David Gow <davidgow@google.com>
Cc:     Brendan Higgins <brendanhiggins@google.com>,
        Shuah Khan <skhan@linuxfoundation.org>,
        Luis Chamberlain <mcgrof@kernel.org>,
        Jeremy Kerr <jk@codeconstruct.com.au>,
        linux-modules@vger.kernel.org, kunit-dev@googlegroups.com,
        linux-kselftest@vger.kernel.org,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Jul 12, 2022 at 5:52 PM David Gow <davidgow@google.com> wrote:
>
> The new KUnit module handling has KUnit test suites listed in a
> .kunit_test_suites section of each module. This should be loaded when
> the module is, but at the moment this only happens if KUnit is built-in.
>
> Also load this when KUnit is enabled as a module: it'll not be usable
> unless KUnit is loaded, but such modules are likely to depend on KUnit
> anyway, so it's unlikely to ever be loaded needlessly.

This seems reasonable to me.

Question: what happens in this case?
1. insmod <test-module>
2. insmod kunit
3. rmmod <test-module>

I think on 3, we'll call the cleanup code, __kunit_test_suites_exit(),
for <test-module>, I think?
But we never called __kunit_test_suites_init().
My fear is what breaks as a result of this precondition break.

E.g. In the case that CONFIG_KUNIT_DEBUGFS is enabled, this includes a
call to kunit_debugfs_destroy_suite() with no previous call to
kunit_debugfs_create_suite().
That will include a call to debugfs_remove_recursive(suite->debugfs),
where suite->debugfs is an uninitialized pointer.

Maybe we can treat it as "undefined behavior" for now and proceed with
this patch.

In terms of long-term fixes, perhaps insmod kunit could trigger it to
1. run all built-in tests (IIUC, it doesn't right now)
2. run all the tests of currently loaded modules
3. track which modules already ran so if you rmmod + insmod kunit
again, it won't rerun tests?

Daniel