Re: Changes to make /dev/*random better sooner

[Thor Lancelot Simon <tls%panix.com@localhost>]

On Wed, Apr 09, 2014 at 08:30:31AM +0200, Manuel Bouyer wrote:
> On Tue, Apr 08, 2014 at 10:33:32PM -0400, Thor Lancelot Simon wrote:
> > > > -               rnd_add_uint32(&skewsrc, rnd_counter());
> > > > -               callout_schedule(&skew_callout, hz);
> > > > +               rnd_add_uint64(&skewsrc, rnd_counter());
> > > > +               callout_schedule(&skew_callout, hz / 10);
> > > 
> > > Do we really need to run 10 extra callouts per second to get an extra bit?
> > 
> > Are you seriously telling me that you are concerned that adding 9 callouts
> > per second will measurably impact system performance?
> 
> The problem I see with this is with power, especially for embeeded systems.
> When we'll go with a tickless kenrel (which I still hope to see), this is
> going to wake up the CPU 10 times per second on an idle system.

When we have a tickless kernel, we can use the polled-entropy-source
functionality I added to make this run only when entropy is urgently
needed.  I'll be glad to make this change as soon as we have a tickless
kernel.