Re: [PATCH v5 2/2] selftests/sgx: Migrate to kselftest harness

[Jarkko Sakkinen <jarkko@kernel.org>]

On Tue, May 18, 2021 at 01:07:16PM -0700, Reinette Chatre wrote:
> Hi Jarkko,
> 
> On 5/18/2021 12:57 PM, Jarkko Sakkinen wrote:
> > On Tue, May 18, 2021 at 08:49:00PM +0300, Jarkko Sakkinen wrote:
> > > On Mon, May 17, 2021 at 10:03:42AM -0700, Reinette Chatre wrote:
> > > > Hi Jarkko,
> > > > 
> > > > On 5/12/2021 2:53 PM, Jarkko Sakkinen wrote:
> > > > > Migrate to kselftest harness. Use a fixture test with enclave initialized
> > > > > and de-initialized for each of the existing three tests, in other words:
> > > > > 
> > > > > 1. One FIXTURE() for managing the enclave life-cycle.
> > > > > 2. Three TEST_F()'s, one for each test case.
> > > > > 
> > > > > This gives a leaps better reporting than before. Here's an example
> > > > > transcript:
> > > > > 
> > > > > TAP version 13
> > > > > 1..3
> > > > > 
> > > > > ok 1 enclave.unclobbered_vdso
> > > > > 
> > > > > ok 2 enclave.clobbered_vdso
> > > > > 
> > > > > ok 3 enclave.clobbered_vdso_and_user_function
> > > > > 
> > > > > Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>
> > > > > ---
> > > > > 
> > > > > v5:
> > > > > * Use TH_LOG() for printing enclave address ranges instead of printf(),
> > > > >     based on Reinette's remark.
> > > > 
> > > > Thank you for considering my feedback. The motivation for my comment was to
> > > > consider how this test output will be parsed. If these tests will have their
> > > > output parsed by automated systems then it needs to conform to the TAP13
> > > > format as supported by kselftest.
> > > > 
> > > > In your latest version the output printed during a successful test has been
> > > > changed, using TH_LOG() as you noted. From what I can tell this is the only
> > > > output addressed - failing tests continue to print error messages (perror,
> > > > fprintf) without consideration of how they will be parsed. My apologies, I
> > > > am not a kselftest expert to know what the best way for this integration is.
> > > > 
> > > > Reinette
> > > 
> > > It's a valid question, yes.
> > > 
> > > The problem is that only main.c can use kselftest macros because
> > > kselftest_harness.h pulls
> > > 
> > > static int test_harness_run(int __attribute__((unused)) argc,
> > > 			    char __attribute__((unused)) **argv)
> > > 
> > > which will not end up having a call site (because there's no
> > > "TEST_HARNESS_MAIN").
> > > 
> > > The whole logging thing in kselftest harness is a bit ambiguous.
> > > Namely:
> > > 
> > > 1. There's a macro TH_LOG() defined in kselftest_harness.h, which
> > >     "internally" uses fprintf().
> > > 2. There's an inline function ksft_print_msg() in kselftest.h
> > >     using vsprintf().
> > > 
> > > To add to that, kselftest_harness.h internally prints by using
> > > ksft_print_msg(), and provides TH_LOG(), which does not use
> > > ksft_print_msg().
> > > 
> > > I don't really get the logic in all this.
> > 
> > I tried to split TH_LOG() as separate entity but it's not possible, as the
> > macros access a static variable called '_metadata'.
> > 
> > I'm not exactly sure how to proceed from this, if we want to make logging
> > consistent.
> > 
> > I would personally suggest to leave the error messages intact in load.c,
> > because there is no way to make them consistent, except by removing them.
> 
> 
> It is not clear to me why ksft_print_msg() cannot be used but an alternative
> to it may be to just prefix all existing diagnostic messages with "# ".
> 
> Reinette

How is using ksft_print_msg() better than using fprintf()? It's as
incompatible with the logging used by fixtures, as is a raw fprintf().

What is the gain of doing that?

/Jarkko