From: Bill Frantz <frantz@netcom.com>
Date: Tue, 21 Oct 1997 15:11:53 +0800
To: cypherpunks@Algebra.COM
Subject: Re: Interactive Programming - Self-modifying code
In-Reply-To: <199710201628.LAA08124@einstein.ssz.com>
Message-ID: <v0300782bb071f9608678@[207.94.249.37]>
MIME-Version: 1.0
Content-Type: text/plain



At 5:24 PM -0700 10/20/97, phelix@vallnet.com wrote:
>I would think that self modifying code would play havoc with today's
>pipelined and superscalar processors (if s&m code would even work on such
>beasts).  Then they're those OS's that won't allow you to muck with code
>space.  However, I guess it will always have a place in the embedded world.

All OSes have a form of self modifying code which fetches programs (as
data) from disk and then executes them (as programs) in memory.  They need
a way to flush the data cache so the instruction cache can read it.

The place where it becomes expensive in modern processors is when the
processor has seen a word as an instruction, and then you want to change it
and have the new value again interpreted as an instruction.  The way to
"legitimately" do this varies by processor.  IBM mainframes detect it
automagically and "do the right thing".  SPARCs require a special "flush"
instruction to clean out the processor I-cache/pipeline.


-------------------------------------------------------------------------
Bill Frantz       | Internal surveillance      | Periwinkle -- Consulting
(408)356-8506     | helped make the USSR the   | 16345 Englewood Ave.
frantz@netcom.com | nation it is today.        | Los Gatos, CA 95032, USA