Gernot Heiser

Not sure I have been following all this correctly, but any sort of cache management is a privilege. If anyone could do it, they can impact performance of other threads,...

If we're talking config options, wouldn't it also make sense to have the option to implicitly ACK the IRQ when waiting on the next one, saving one syscall?

> With an edge-triggered interrupt you need to ACK the IRQ before handling the interrupt, or at least double check there wasn't a new event after ACKing. If you don't,...

In practice, 64KiB pages are much more useful than 2MiB pages, not only for the reasons @kent-mcleod mentions, but also for performance optimisation.

colouring with superpages doesn't seem likely. If you have a 1MiB L2, then with 64KiB pages, your L2 associativity cannot be higher than 8 to get at least 2 colours....

Can the kernel check what prio is running on the destination core? It would be good to only interrupt if the migrating thread should actually preempt what's currently running

On 10 Feb 2022, at 10:02, Gerwin Klein ***@***.***> wrote: > > Happy with the waiting as such, just wanted to check on the way it's done. @kent-mcleod might know...

Can’t look at the code right now (very poor connection) but from memory a lot of data is written by the benchmarking rig to a buffer to collect statistics, which,...

Call is only one side of an IPC. A round-trip is Call+Reply, which at the moment is 383+389=772 cycles.