From: Claudio Jeker Subject: Re: cpu perfpolicy To: Theo de Raadt Cc: Rafael Sadowski , Vitaliy Makkoveev , Ted Unangst , tech@openbsd.org Date: Tue, 3 Jun 2025 20:26:24 +0200 On Tue, Jun 03, 2025 at 07:29:57AM -0600, Theo de Raadt wrote: > Rafael Sadowski wrote: > > > On Mon Jun 02, 2025 at 05:37:00PM +0300, Vitaliy Makkoveev wrote: > > > On Sun, Jun 01, 2025 at 03:22:59PM -0400, Ted Unangst wrote: > > > > This introduces a helper function for hw.perfpolicy that allows more > > > > machine specific policies to be set. > > > > > > > > On my laptop, there's also a fan policy. This affects performance, but > > > > it's not really related to the existing hw.setperf mechanism. For > > > > example, a "silent" fan setting and "high" CPU setting is reasonable. > > > > Exactly what I want, in fact. > > > > > > > > Fortunately, perfpolicy takes a string, which makes it flexible. > > > > This uses two new optional functions to parse and append the policy. > > > > > > > > hw.perfpolicy=silent,auto > > > > > > > > > > Do we really need to combine them? Why don't have dedicated > > > hw.fanpolicy? > > > > > It continues to be very weird to demand more "root has to do manual fiddling" > in a subsystem which advertises itself as being all about automated management. > > How many people will set these control knobs? Will it be 3 people, or will > it be 4 people? Most people want their machine to be as fast as possible. While I want a machine that is as fast as possible I also want it to be quiet. On some machines you need to twiddle with these knobs or hack the kernel to get a system that works optimal for the user. -- :wq Claudio