Hi Peter,
On 01/30/2014 10:37 AM, Norman Feske wrote:
Hello Peter,
It seems like Genode's secure display windows are off to a goodstart with "Nitpicker"... ...What is happening so that the Wayland display compositor will be able to in some way run withing the Genode OS?
as far as I know, there is no concrete plan to bring Wayland to Genode. It is not mentioned on our road map [1]. However, we have listed the topic at our "challenges" page as a rough idea.
[1] http://genode.org/about/road-map [2] http://genode.org/about/challenges
Design-wise there seem to be many similarities between nitpicker and Wayland, but both solutions were created under different premises. For Wayland, fluent graphics without any tearing artifacts were a big motivation. The goal was to overcome the deficiencies of X.org. In contrast, nitpicker was designed for security in the first place (you can read more details in my dissertation [3]). Both goals led to a similar design, which is a very good sign.
[3] http://www.genode-labs.com/publications/secure-gui-2009.pdf
Personally, I think that exploring the combination of Wayland with Genode is an interesting topic, but I am anything but sure about the outcome. Whereas Wayland was designed for Linux, it is unclear to me how well it fits with Genode's capability-based security model. Also, a Wayland-based graphics stack will be significantly more complex than a nitpicker-based one. This is because Wayland uses OpenGL as backend. So Mesa plus the whole driver stack becomes a mandatory part of the trusted computing base for any graphical application. In contrast, nitpicker's trusted computing base is orders of magnitude less complex.
That said, I think this does not need to be an either-or discussion because different users have different needs. Not everyone is picky about a low-complexity trusted-computing base. Having the option to use Wayland or nitpicker would be great.
As a technical precondition to take a closer look at Wayland on Genode, we first need to address the problem to bring our version of Mesa up to date and provide a way to use hardware-accelerated graphics. Those topics are also important for Qt5's QML. For this reason, I raised the point during the discussion of the road map. Even though we haven't put it on the official road map, we still plan to work on it.
Where might I find more info on the efforts to run Genode within ARMTrustZone...
We are positively surprised about the response to our TrustZone work. Currently, there is not much documentation available. But we will obviously need to change that.
For practical steps of how to start experimenting with TrustZone, maybe Stefan can give you a good starting point?
Btw, there will be talk by him at FOSDEM on Sunday in Brussels. We will publish the slides and there may be even a video recording.
Concurrently, we've made just some experiments regarding ARM's TrustZone. The farthermost developed demonstration scenario therefore is build for Freescale's i.MX53 SABRE tablet. The code for this scenario isn't completely part of Genode's mainline code, but can be found here:
https://github.com/skalk/genode/tree/i.MX53_tablet_demo
A brief "how to" reproduce that scenario is included in there too:
https://github.com/skalk/genode/blob/i.MX53_tablet_demo/os/src/server/vmm/im...
The TrustZone experiments are limited to the "base-hw" platform. Most code, including the switch between "secure" and "normal" world, can be found there. I hope this will help as a starting point.
Regards Stefan
Best regards Norman