Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210715213138.1363079-1-dlatypov@google.com> <20210715213138.1363079-2-dlatypov@google.com>
 <20210723064328.GA7986@gondor.apana.org.au>
In-Reply-To: <20210723064328.GA7986@gondor.apana.org.au>
From:   Daniel Latypov <dlatypov@google.com>
Date:   Fri, 23 Jul 2021 12:31:28 -0700
Message-ID: <CAGS_qxqOD+Bcvy7xti7_eg8+H1cJcfp94BtnRhuzijDcaGF_uA@mail.gmail.com>
Subject: Re: [RFC v1 1/2] crypto: tcrypt: minimal conversion to run under KUnit
To:     Herbert Xu <herbert@gondor.apana.org.au>
Cc:     davem@davemloft.net, linux-crypto@vger.kernel.org,
        brendanhiggins@google.com, davidgow@google.com,
        linux-kernel@vger.kernel.org, kunit-dev@googlegroups.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Jul 22, 2021 at 11:43 PM Herbert Xu <herbert@gondor.apana.org.au> wrote:
>
> On Thu, Jul 15, 2021 at 02:31:37PM -0700, Daniel Latypov wrote:
> >> == Questions ==
> > * does this seem like it would make running the test easier?
>
> I don't mind.  tcrypt these days isn't used so much for correctness
> testing.  It's mostly being used for speed testing.  A secondary
> use is to instantiate templates.

Thanks, that makes a lot of sense.
In that case, how useful would `kunit.py run` be? I.e. Do people
mostly want to see numbers on bare metal?

The default mode of `kunit.py run` is to use ARCH=um.
I assume (for at least most of the library-type crypto code) it should
have the same performance characteristics, but that might not be the
case. I can try and get some numbers on that.

There's an option to make `kunit.py run` use ARCH=86_64, but it'll be
in a QEMU VM, so again there's some performance overhead.

If either option seems useful, then perhaps a minimal patch like this
would be beneficial.
I can make it even smaller and less intrusive by restoring the "ret +=
..." code and having a single `KUNIT_EXPECT_EQ_MSG(test, ret, 0, "at
least one test case failed")` at the very end.

It does not sound like patch #2 or any future attempts to try and make
use of KUnit features is necessarily worth it, if correctness testing
isn't really the goal of tcrypt.c anymore.

>
> > * does `tvmem` actually need page-aligned buffers?
>
> I think it may be needed for those split-SG test cases where
> we deliberately create a buffer that straddles a page boundary.
>
> > * I have no clue how FIPS intersects with all of this.
>
> It doesn't really matter because in FIPS mode when a correctness
> test fails the kernel panics.
>
> >   * would it be fine to leave the test code built-in for FIPS instead of
> >   returning -EAGAIN?
>
> The returning -EAGAIN is irrelevant in FIPS mode.  It's more of
> an aid in normal mode when you use tcrypt for speed testing.
>
> Thanks,
> --
> Email: Herbert Xu <herbert@gondor.apana.org.au>
> Home Page: http://gondor.apana.org.au/~herbert/
> PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt