On Sun, Nov 03, 2024 at 10:08:14AM +0100, Martin Pieuchot wrote:
> claudio@ and dlg@ found that the current contention point in the fault
> handler is due to the exclusive locks needed by uvm_fault_lower_lookup().
> I'm working on reducing this contention by using reader version of the
> locks when possible.
> 
> As for parallelizing the fault handler, I'll start by reducing the
> differences with NetBSD's UVM then adapt their changes to our tree.  Diff
> below is a first step in that direction.  It introduces a new function,
> uvm_fault_lower_io(), without behavior change.
> 
> ok?

Obvious refactor, no behavior change.

ok tb