Scott Cheloha writes:
> On Fri, Jun 04, 2021 at 04:02:01AM +0200, Alessandro Pistocchi wrote: >> To give you an idea, I am giving exclusive access to 3 out of 4 cpu cores >> to a game > > Scheduler affinity control is useful in some cases. We don't have it, > but it would be useful, even if only for testing purposes. > One niche area is for users of vmm(4) on Intel VMX-based hosts. Intel's design requires synchronizing state stored on the cpu to physical memory if/when the vmm kernel thread moves cpus. It then needs to be reloaded on the new cpu. (This is an area I'm currently working on to improve as we have some race conditions exposed via some of the unlocking that's been going on.) But do we need it? *shrug* The Intel SDM alludes to performance gains for having a virt-aware scheduler, but stops short of actually giving hard numbers from any OS, so I'm not entirely convinced that they are convinced. -dv
