[wizeman]

Wow, that's so cool!

That's actually much better than what I was thinking and if I understand you correctly, that should allow you to exploit practically all possible parallelism opportunities (given the constraints).

I'm guessing that your plan would require unmapping shared memory pages so that Hermit can catch memory accesses and request permission from the scheduler (to access the memory pages) when these memory accesses happen. But I think it should be possible to do it, yes (with some overhead if two threads/processes access shared memory pages simultaneously very often, I guess).

Awesome work, thank you!