Coincidentally, this recent HN post also contains a fine introduction to processor internals of this sort: https://gamozolabs.github.io/metrology/2019/08/19/sushi_roll.... I know you (BeeOnRope) don't need the introductory parts, but you might appreciate the meat of his post.
Yes, I've read and it have been thinking about how to do this kind of synchronized counter reading from userspace, where it would be easier to access (even if you wanted to go the kernel route, this kernel is not publicly available).
I really like this way of displaying per-cycle information from the performance counters. Very cool.
