Subject: Re: CVS commit: src/sys/arch
To: None <source-changes@NetBSD.org>
From: Joerg Sonnenberger <joerg@britannica.bec.de>
List: source-changes
Date: 10/26/2007 15:47:29
On Fri, Oct 26, 2007 at 02:28:52PM +0100, Andrew Doran wrote:
> A DIAGNOSTIC kernel should probably panic on a very long delay..

I was going to ask if we want to introduce a reasonable upper limit.
IMO anything above a few second is a major bug in the kernel.

Another question I would raise is whether we want to bother with the
delaytab in the current size. Based on my old documeentation, a i386
would need less then cycles to compute the delay and for a base of
25MHz, we can compute delay(4) and above directly. For anything modern,
I would go as far as wire down only 0 and 1. Simon measured around 0.2us
on a Pentium MMX, which supports this thesis.

The same argument applies to lapic, which I haven't touched yet. It
needs some more work as it uses 64bit multiplications more often, but
the idea applies as well.

Joerg