Roadmap 2018

[Norman Feske]

Hi Stefan,

> To me personally it seems 2018 will be the year of "sculpt", and
> although I much appreciate my daily work on a Genode base, I fear
> other activities will become orphaned - which is our own kernel and
> support of embedded platforms.

Let me try to dispel your fear. My line of thoughts is as follows:

1. NOVA cannot be Genode's long-term future.

We carry the burden of maintaining NOVA alone (where "we" is for the
most part one person - hi Alex!). Note that I am speaking of
maintaining, not developing. It is clear that we won't expand our
engagement beyond the current level. Solving the remaining fundamental
deficiencies of the kernel, in particular the kernel-resource
management, is not on our agenda. This would be a different story if
there was an active community around the kernel. But there is none to
speak of.

Feature-wise, NOVA is complete. So it is definitely a suitable basis for
today and the immediate future. But in the longer term, its limitations
will become a burden that we must overcome.

2. Base-hw solves NOVA's architectural problems but lacks features.

The base-hw kernel facilitates Genode's architecture to solve problems
that are extremely complicated to solve in other kernels (seL4 comes to
mind) or unsolved (NOVA). But its feature set is incomplete.

Since we won't take NOVA to the level we need in the longer term, the
conclusion can only be what you just suggested: adding the missing
features to base-hw:

> * revert kernel optimization to share the address space with other
>   components at least on top of x86, keyword: meltdown
> * enable SMP for x86
> * implement IOMMU support
> * add x86 hypervisor functionality to hw

So how about combining the overall theme "Sculpt" with this ambition?
Should we set up the goal to run Sculpt on base-hw by the end of the year?

You may wonder about the role of seL4 in this discussion. In contrast to
NOVA, it is actively developed and there is a welcoming community around
it. It is also fairly feature complete. On the other hand, it is
complicated and we cannot easily tailor it to our needs. This is not a
critique but a just consequence from seL4's focus. In contrast to seL4,
base-hw was co-designed with Genode. So it is simple and frictionless,
and puts us in control over everything. Hence, I see seL4 and base-hw in
complimentary roles. (A) In situations where seL4's formal-verification
story is anticipated, Genode enables seL4 to carry highly dynamic
workloads. (B) For Genode use cases that call for the highest-possible
simplicity, or that benefit from our full control over the kernel,
base-hw is attractive.

Even though I often hear that (A) is super interesting, Genode/seL4 has
no commercial backing yet. I.e., there are no customers with the
ambition to drive Genode's seL4 support via commissioned development
projects. Naturally, as a Genode developer, my personal use case falls
into category (B). So I wholeheartedly support your plan!

Cheers
Norman

-- 
Dr.-Ing. Norman Feske
Genode Labs

https://www.genode-labs.com · https://genode.org

Genode Labs GmbH · Amtsgericht Dresden · HRB 28424 · Sitz Dresden
Geschäftsführer: Dr.-Ing. Norman Feske, Christian Helmuth