Hello Chris,
On Wed, May 30, 2018 at 12:11:03PM -0400, Chris Rothrock wrote:
Working with our developers for SABLE, we branched a build of SABLE for one purpose - to NOT run the AMD skinit instruction before handing off boot to NOVA/Genode. Using this build of SABLE we are able to get a successful boot to the demo.
At this stage, it seems that something fundamental changed within Genode and/or NOVA (and I have also seen this with Genode/seL4 demo) that causes the memory fault to occur on AMD platforms ONLY when the skinit instruction is performed prior to boot hand-off. Is this a point where we can dig a little deeper to understand where this problem is occurring?
Your issue description suggests you have the perfect "git bisect" situation: an old source-code revision where it seems to work and a new revision that shows an issue. Therefore, I recommend you try to narrow down the time frame where the issue appears first by testing different revisions of Genode with a minimalistic scenario like "make run/log".
From our side I can't promise any practical contribution as the
current release works pretty stable on our test hardware (admittedly Intel not AMD) and we don't have access to SABLE.
Regards