On Mon, Apr 26, 2021 at 12:57:33PM +0200, Mark Kettenis wrote: > > I did measure kernel build time make -j 4 on a 4 core machine. > > System time goes down by 7.7%, real time stays the same. > > That basically suggests that contention of the kernel lock isn't the > major bottleneck for this workload.
Kernel lock is a bottleneck, but the diff does not solve it completely. > > http://bluhm.genua.de/files/kstack-make.svg > > http://bluhm.genua.de/files/kstack-make-uvm.svg In current we have 25% usertrap -> upageflttrap -> kernel_lock. With diff we have 22% usertrap -> upageflttrap -> uvm_fault -> kernel_lock. So it pushes the problem down the call stack. bluhm