Steve Reinhardt wrote: > I don't think devices do anything on functional accesses; you probably > want an atomic access. That shouldn't be a problem as long as the > memory system is not in timing mode. Is this just for kvm > initialization or does it occur during normal operation?
It does this during normal operation for any memory address you haven't explicitly set up as physical memory. It's happening right now because I left a hole for the VGA framebuffer like QEMU apparently does, and I think the kernel is clearing it out as part of early initialization. I haven't actually looked at what's at the IP of interest, though, so it might be doing something else. The same thing would happen for any/all IO accesses since they use a very similar model and become only slightly tweaked memory accesses anyway. > > Is it possible just to make the access just look like it's coming from > cpu0? That seems easiest to me. Yes, and that's what I was thinking I might end up doing. I didn't want to do it that way if there was already a cleaner mechanism, but I can understand why there hasn't been a need for that before. Gabe _______________________________________________ m5-dev mailing list m5-dev@m5sim.org http://m5sim.org/mailman/listinfo/m5-dev
