Re: [RFC v3 06/19] arch: um: enable running kunit from User Mode Linux

[Brendan Higgins]

On Fri, Nov 30, 2018 at 10:22 AM Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
>
> On Fri, Nov 30, 2018 at 08:05:34AM -0600, Rob Herring wrote:
> > On Thu, Nov 29, 2018 at 9:37 PM Luis Chamberlain <mcgrof@xxxxxxxxxx> wrote:
> > >
> > > On Wed, Nov 28, 2018 at 03:26:03PM -0600, Rob Herring wrote:
> > > > On Wed, Nov 28, 2018 at 1:37 PM Brendan Higgins
> > > > <brendanhiggins@xxxxxxxxxx> wrote:
> > > > >
> > > > > Make minimum number of changes outside of the KUnit directories for
> > > > > KUnit to build and run using UML.
> > > >
> > > > There's nothing in this patch limiting this to UML.
> > >
> > > Not that one, but the abort thing segv thing is, eventually.
> > > To support other architectures we'd need to make a wrapper to that
> > > hack which Brendan added, and then allow each os to implement
> > > its own call, and add an asm-generic helper.

I think Rob is referring to the description for this patch. This patch
previously did what you suggested, Luis, (source the KUnit kconfig
from arch/um/) but Kees asked me to change it to how it is now (which
probably makes sense if we are saying KUnit is not intended to be tied
to a particular architecture, no?), and I forgot to update the commit
description, sorry.

> >
> > I've not looked into why this is needed, but can't you make the abort
> > support optional and arches can select it when they support it.
>
> Its why I have asked for it to be properly documented. The patches in no
> way illustrate *why* such thing is done. And if we are going to
> potentially have other archs do something similar best to make it
> explicit.

Yeah, I should better document it. I should also probably not include
any UML specific header files in kunit/test.h; that seems like I am
asking to get more tightly coupled if I am not careful about exactly
what things I depend on.

I think Luis is right, I need to add a wrapper around the features
needed for the hack to support abort() and then write a UML specific
implementation.

For the asm-generic case, we could probably just have abort() call
BUG(), with that KUnit should work on most architectures, albeit with
pretty reduced functionality.

>
> > At
> > least before, the DT unittests didn't need this to run and shouldn't
> > depend on it after converting to kunit.

Fair enough.