Re: [PATCH v2 0/5] crypto: hisilicon - add HiSilicon SEC V2 support

[Eric Biggers <ebiggers@kernel.org>]

On Tue, Nov 12, 2019 at 10:11:10AM +0800, Xu Zaibo wrote:
> 
> On 2019/11/12 9:28, Eric Biggers wrote:
> > On Tue, Nov 12, 2019 at 09:04:33AM +0800, Xu Zaibo wrote:
> > > On 2019/11/12 1:18, Eric Biggers wrote:
> > > > On Mon, Nov 11, 2019 at 08:26:20PM +0800, Xu Zaibo wrote:
> > > > > Hi,
> > > > > 
> > > > > On 2019/11/11 13:37, Eric Biggers wrote:
> > > > > > On Mon, Nov 11, 2019 at 10:21:39AM +0800, Xu Zaibo wrote:
> > > > > > > Hi,
> > > > > > > 
> > > > > > > On 2019/11/9 10:16, Eric Biggers wrote:
> > > > > > > > On Sat, Nov 09, 2019 at 10:01:52AM +0800, Zaibo Xu wrote:
> > > > > > > > > This series adds HiSilicon Security Engine (SEC) version 2 controller
> > > > > > > > > driver in Crypto subsystem. It includes PCIe enabling, Skcipher, DebugFS
> > > > > > > > > and SRIOV support of SEC.
> > > > > > > > > 
> > > > > > > > > This patchset rebases on:
> > > > > > > > > git://git.kernel.org/pub/scm/linux/kernel/git/herbert/cryptodev-2.6.git
> > > > > > > > > 
> > > > > > > > > This patchset is based on:
> > > > > > > > > https://www.spinics.net/lists/linux-crypto/msg43520.html
> > > > > > > > > 
> > > > > > > > > Changes:
> > > > > > > > >      - delete checking return value of debugfs_create_xxx functions.
> > > > > > > > > 
> > > > > > > > > Change log:
> > > > > > > > > v2:    - remove checking return value of debugfs_create_xxx functions.
> > > > > > > > > 
> > > > > > > > Does this driver pass all the crypto self-tests, including with
> > > > > > > > CONFIG_CRYPTO_MANAGER_EXTRA_TESTS=y?
> > > > > > > > 
> > > > > > > Not including extra testing now, only CONFIG_CRYPTO_TEST is passed.
> > > > > > > 
> > > > > > Can you please ensure that all the extra tests are passing too?  I.e., boot a
> > > > > > kernel with CONFIG_CRYPTO_MANAGER_EXTRA_TESTS=y and check dmesg for failures.
> > > > > > 
> > > > > Ok, I will try to do this. BTW, why we need this test? Thanks.
> > > > > 
> > > > It will test the correctness of your driver.
> > > > 
> > > So, it is a basic test not an extra test ? :)
> > > 
> > The options are separate because the "extra tests" include fuzz tests which take
> > much longer to run than the regular tests, and some people who enable the
> > regular tests wouldn't want them to get 100x slower.  But as someone actually
> > developing a crypto driver you're expected to run the extra tests.  They've
> > found lots of bugs in other drivers, so please run them and fix any bugs found.
> > 
> Okay. Not sure whether my understanding is right. Should it be a part of
> regular test once
> we find a shorter time way to do this "extra tests"?
> Yes, I am running it. Thank you very much.

You should always run them when doing crypto development.  With the default
fuzz_iterations=100 they normally just take a few seconds.  It's just that not
everyone who wants to enable some level of crypto testing is actively developing
a crypto driver and is willing to delay boot by any significant amount, so we
have to have multiple testing options.

- Eric