From: Vitaliy Makkoveev <mvs@openbsd.org>
Subject: Re: kevent(2): make EVFILT_TIMER mp-safe
To: Theo de Raadt <deraadt@openbsd.org>
Cc: Ted Unangst <tedu@tedunangst.com>, Visa Hankala <visa@hankala.org>, tech@openbsd.org
Date: Thu, 8 May 2025 21:12:23 +0300

On Thu, May 08, 2025 at 11:38:34AM -0600, Theo de Raadt wrote:
> Vitaliy Makkoveev <mvs@openbsd.org> wrote:
> 
> > On Thu, May 08, 2025 at 01:24:24PM -0400, Ted Unangst wrote:
> > > On 2025-05-08, Vitaliy Makkoveev wrote:
> > > > -int kq_ntimeouts = 0;
> > > > -int kq_timeoutmax = (4 * 1024);
> > > > +int kq_ntimeouts = 0;		/* [a] */
> > > > +int kq_timeoutmax = (4 * 1024);	/* [I] */
> > > 
> > > Unrelated to this diff, but it would be nice to make these limits per
> > > process, like with the user event diff. (I still like the user events.)
> > 
> > Yes, it will be better.
> 
> Won't processes ask for the most system-punishing (highest cost) timeouts?
> 

They already do. This diff only scales the pressure on existing CPUs. I
have positive reports for previous iteration of this diff.