Received-SPF: pass (google.com: domain of linux-kernel+bounces-212132-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Date: Wed, 12 Jun 2024 12:01:27 -0700
In-Reply-To: <20240612104500.425012-2-usama.anjum@collabora.com>
Precedence: bulk
Mime-Version: 1.0
References: <20240612104500.425012-1-usama.anjum@collabora.com> <20240612104500.425012-2-usama.anjum@collabora.com>
Message-ID: <Zmnwhx0Y0qh0x03J@google.com>
Subject: Re: [PATCH 2/2] selftests: kvm: replace exit() with ksft_exit_fail_msg()
From: Sean Christopherson <seanjc@google.com>
To: Muhammad Usama Anjum <usama.anjum@collabora.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>, Shuah Khan <shuah@kernel.org>, 
	Anup Patel <anup@brainfault.org>, Oliver Upton <oliver.upton@linux.dev>, 
	Claudio Imbrenda <imbrenda@linux.ibm.com>, kernel@collabora.com, kvm@vger.kernel.org, 
	linux-kselftest@vger.kernel.org, linux-kernel@vger.kernel.org
Content-Type: text/plain; charset="us-ascii"

On Wed, Jun 12, 2024, Muhammad Usama Anjum wrote:
> The KSFT_FAIL, exit code must be used instead of exit(254).

This needs more justification.  KVM selftests have worked just fine for 6+ years
using exit(254), so stating they "must" use KSFT_FAIL is obviously not true.

I'm not personally opposed to switching to KSFT_FAIL, but it is a potentially
breaking change.  E.g. some of Google's internal test infrastructure explicitly
relies on the exit code being 254.  I don't _think_ that infrastructure interacts
with KVM selftests, nor do I think that forcing upstream KVM selftests to sacrifice
TAP compliance just to play nice with someone's crusty test infrastructure is a
good tradeoff, but this and all of the TAP compliance work needs to be done with
more thought and care.

> The 254 code here seems like anciant relic.

As above, AFAICT it comes from Google's internal test infrastructure (KVM selftests
came from Google).

> Its even better if we use ksft_exit_fail_msg() which will print out "Bail
> out" meaning the test exited without completing. This string is TAP protocol
> specific.

This is debatable and not obviously correct.  The documentation says:

  Bail out!
  As an emergency measure a test script can decide that further tests are
  useless (e.g. missing dependencies) and testing should stop immediately. In
  that case the test script prints the magic words

which suggests that a test should only emit "Bail out!" if it wants to stop
entirely.  We definitely don't want KVM selftests to bail out if a TEST_ASSERT()
fails in one testcase.

> Signed-off-by: Muhammad Usama Anjum <usama.anjum@collabora.com>
> ---
>  tools/testing/selftests/kvm/lib/assert.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/tools/testing/selftests/kvm/lib/assert.c b/tools/testing/selftests/kvm/lib/assert.c
> index 33651f5b3a7fd..db648a7ac429b 100644
> --- a/tools/testing/selftests/kvm/lib/assert.c
> +++ b/tools/testing/selftests/kvm/lib/assert.c
> @@ -87,7 +87,7 @@ test_assert(bool exp, const char *exp_str,
>  
>  		if (errno == EACCES)
>  			ksft_exit_skip("Access denied - Exiting\n");
> -		exit(254);
> +		ksft_exit_fail_msg("\n");
>  	}
>  
>  	return;
> -- 
> 2.39.2
>